Regardless of whether they achieved parity via distillation, or whether they got here via independently constructing a model from scratch, it was always going to end this way for the frontier American labs. Distillation “attacks” are not attacks. The frontier labs “distilled” all existing human written knowledge into their models, there was always going to be a second class lab that would distill that model into a cheaper version of it. There was never any plausible explanation for why this wouldn’t happen. There was never any practical mechanism to prevent someone from saving a conversation and using it to train their own model.
Even if it didn’t happen here, it was still the case that it was going to happen going forward. It was always going to end like this. Invest in the hardware companies, not the model companies.
I strongly agree with the premise that distillation is not an “attack”.
But that said: K3 is not a distilled version of Fable or Sol. Fable has been barely available and Sol was just released! Moreover, K3 is superior to both models in some domains, according to user scoring on the Arena.
API distillation can’t give you these results anyway. All it is useful for is bootstrapping RL in new domains to get past the “cold start” problem faster. By far, what matters more is the quality and variety of RL environments the model learns from.
This analysis observed K3 identifies itself as Claude approximately 15% of the time.
K3 reproduces Claude's correct current model id, which the real Claude models themselves do not emit. This suggests K3 was trained on Claude data labeled with deployment metadata (API logs, tagged synthetic data), rather than Claude's chat outputs.
Kimi calling itself claude means nothing. During pre-training, when the model learns to "simulate" the internet text, it will naturally be fed with a bunch of data about Claude and ChatGPT. With the amount of LLM outputs on the internet today, it is not surprising at all that a model would naturally call itself Claude or ChatGPT. You can mitigate that in post-training (or actually in pre-training as well) by training on many examples of what the model should call itself. That being said, getting probably hundreds pf thousands of ChatGPT and Claude examples totally "pirged" out of the weights is going to be difficult and really more hassle than its worth.
Sure, but then Qwen should leak that too, and it doesn't. K3 calls itself Claude 7 out of 48 times, Qwen does it 0 out of 48, and the only other model to identify itself as Claude is DeepSeek. and DeepSeek is alleged to also distill from Claude data anyway. So this isn't something every model absorbed from the same web text.
And you skipped over the strongest datapoint that K3 is distilled: K3 reproduces Claude's public model identifier under prefill (i.e. "claude-opus-4-5-20251101"). This data does not appear in Claude chat logs, only in API logs. K3 only does this for Claude models and not for any other lab. The real Claude models don't produce their own current public identifier, they only know their previous identifier (i.e. Sonnet 4.5 calls itself "Claude 3.5 Sonnet").
This is highly suggestive of the type of data that K3 was trained on. K3 was very likely trained on Claude metadata traces (API logs, tagged synthetic data). Not web chat logs, those wouldn't include this identifier. And this data wasn't filtered correctly, which is why K3 incorrectly identifies itself as Claude 15% of the time.
> Kimi K3 reproducibly identifies itself as Claude
It could also be have been trained from collected response datasets. Claude got caught several time responding it was ChatGPT or even Deepseek and I don't think Anthropic has been distealling DeepSeek.
> This behavior is exactly what you'd expect from a model distilled from Claude.
The opposite actually. If they wanted to distill Claude without getting caught they could just use a regex to change Claude to Kimi in their distillation pipeline!
Though I am of the opinion that distilling is no different than how extant frontier LLMs have also been trained on other people's data, I could actually see the word distealling becoming useful in discussion.
Its not a typo, someone coined that during the DeepSeek R1 hype period and I kept using it since then.
I totally agree with you on the fact that it's not morally any different than pre-training. IMHO we should have a legislation that force base models to be released publicly without any restrictions whatsoever as it's basically the product of the whole humanity's intelligence.
Surprising they didn't clean that from the data before training. It's easy to identify, a simple search->replace gets most of it, and a cheap LLM can identify the edge cases (e.g. avoiding "Claude Shannon" -> "Kimi Shannon" or something).
nothing new, all ai labs are immoral and not bound by any reasonable oversight or ethical constraints. All outlaws in their own rights on that front. Absolutely none of them have true rights on the matter of being distilled from given historic and continued behaviour. I'm not sure why this is a talking point at all? We know AI companies steal, the least interesting behaviour among this is them stealing from one another.
For me, a far more interesting and important point of conversation on this matter is anthropic buying rare or evwn unique books, processing them for training data, and then destroying the books for others cannot use it as well.
Permanemt destruction of priceless primary source materials is so many leagues beyond copying a copy that I cannot fathom it even registering as a discussion point.
> For me, a far more interesting and important point of conversation on this matter is anthropic buying rare or evwn unique books, processing them for training data, and then destroying the books for others cannot use it as well.
That's an incredible allegation, and appalling if true. But is it true?
But in my opinion, treating mass produced books like they're this sacred untouchable object is ridiculous. They're not "source" material, they're just a copy as well, and they're not "priceless" by any means. They're very reasonably priced, perhaps even so cheaply priced that books can be bought in bulk in these amounts. Buying used books and doing whatever you want with them is just legal. Used books, that would probably be just laying in some warehouse, or recycled anyway.
If there's anything to have gripes with, it's the copyright system that makes it easier to take this legal route.
Your main source is Ryan Greenblatt who is a regular recipient of community notes and has no corroboration for the 15% statistic other than his assertion. The other tweet (Sauers_) is also community noted as engagement farming with a false system prompt, so forgive me for being skeptical.
The main story is what isn't being talked about. Chinese labs exfiltrated trillions of tokens of high-quality output from Anthropic and OpenAI, through proxies and heavily discounted token resellers, which they distilled and used for training data for their own models.
Instead of spending 12-18 months building their own robust harnesses and painstakingly creating quality training data (which is what Anthropic and OpenAI did), they distilled Anthropic's models to bypass the hardest parts of development. Chinese labs compressed 18 months of intensive research and development into just 6 months, and are now head-to-head with their American counterparts.
Anthropic tried to complain about this unauthorized "token theft", but they burned too much public goodwill with BS safety restrictions and users don't care. The US government is too busy fighting a war to help. Chinese labs are offering highly capable, cheap, open-weight models; exactly what users want. The community is happy to overlook any questionable methods Chinese labs used to build them.
The cope is incredible. There's people in this thread in denial that Moonshot AI is trained on exfiltrated Anthropic's model output, even when shown substantial evidence this has been happening since Kimi 2.X
Chinese labs were even paying an absurd $0.01 per Opus tool call trace, to get the quantity of training data needed.
Kimi K3 has reached the point of RSI, and no longer needs synthetic data generated by Anthropic/OpenAI models. K3 is now capable enough to generate, iterate, and improve its own training data recursively. The data exfiltration is complete.
We witnessed the most extensive industrial espionage campaign, probably ever, and nobody in the industry cares at all that it happened.
I could perhaps get myself to care just the tiniest bit if the information that was supposedly stolen wasn't generated by "stealing" from everybody else. Either it is fair use to train AI models on whatever information you can get your hands on for everyone or for no one.
Ask claude its name in Chinese and it says Qwen or Deepseek. Anthropic distilled Chinese tokens rather than create their own Chinese language training data.
Hard to feel sorry for companies that created their empires by ignoring copyright themselves.
Also, 'most extensive industrial espionage campaign, probably ever' is absolute nonsense. They did not need to infiltrate the companies for this nor are you accusing them of stealing any trade secrets. This is only about whether they looked at their competitors' products from the outside (in the form of conversation tokens) and used it to improve their own product (by training). Hardly the crime of the century.
You are completely underestimating the scale of what is happening here.
Chinese AI labs are actively facilitating an industrial-scale network of tens of thousands of bot accounts, that resell Claude tokens at 97% below official API prices. They buy subsidized Max 5x plans (sometimes with stolen credit cards), then split the subscription across dozens of clients and reselling the output. They are running a massive data-harvesting operation. Chinese labs and token resellers subsidize the cost of the tokens in exchange for the API metadata (detailed reasoning traces, model outputs, and tool calls) to use as high-quality training data for their own models.
They are buying Anthropic's own product, just to resell it below cost, just so they can capture the training data. Reportedly, they are paying as much as ~$0.01 per tool call.
You haven't explained how this is illegal or any more immoral than scraping the web for training data.
As you said yourself: They are buying the product. Then they are using it for their own purposes. That's more than Anthropic/OpenAI did for the open internet. That's more than Meta did when they obtained torrents of books in the early days, and then claimed that even though the data was obtained illegally they can still train on it just fine.
They paid for it! It's absurd to call this espionage!
They didn't though. The resellers are not buying via the official API, they're buying Max subscriptions (where tokens are priced ~10x below API cost), then splitting the subscription across dozens of clients and reselling the output as the regular API. Anthropic prices its subscription plans barely at cost, to bring in customers onto their enterprise plans where they can charge expensive API rates. Reselling these subsidized plans for price arbitrage is a TOS violation. It's not a legitimate purchase. Plus, a non-trivial amount of this volume is funded by stolen credit cards, so this "revenue" gets chargeback anyway.
The resellers then log all the the model output, then sell it to Chinese labs as training data.
> is it any more immoral than scraping the web for training data.
I think you'd acknowledge there's a difference between "We indexed public web pages" and "We deployed tens of thousands of fraudulent accounts to resell your subsidized plan for cheap, stealing your own customers, while collecting the data to build our own competing product" are very different actions. One can believe the first was wrong while acknowledging the second is far worse.
TOS violations are not espionage. Everybody who links up Claude to OpenCode is violating the TOS.
So from the largest industrial espionage in history we have left "They paid for the accounts but violated the TOS". And then you randomly add the claim they stole the money to pay for the accounts.
You have provided no evidence other than "Claude tokens are sold for cheap in China". As others have pointed out, that might also simply be counterfeit tokens generated by open weight models.
The western labs have established the precedent that all data they can buy beg borrow or steal is fair game. Turning around and crying foul when the Chinese labs follow their lead is hypocrisy.
Have a read of this detailed article, it's well sourced and documented that token resellers are logging the Claude outputs and selling them to Chinese labs. All your points are addressed in there https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
I linked it earlier, but it seems you didn't see it.
First off, I don't doubt that China engages in large-scale industrial espionage, or at least used to. Nowadays they have plenty of talented and highly educated engineering talent of their own.
Second, I now actually read the article. It describes plenty of questionable and problematic things but also contradicts your claims explicitly.
The essential point of the article is about making money by selling access to Claude cheaply in China. Not about Chinese labs orchestrating a way to get their hands on Claude output.
Your credit card claim is considerably weaker in the article: "[Beyond this there are] accounts purchased using stolen or fraudulent credit cards [...]. How large this share is relative to the above four “innocent” tactics is difficult to verify, but the two markets likely share some infrastructure and personnel.". Instead, swapping models to cheaper alternatives is listed as a major reason for cheaper prices.
Then the article gets a key point wrong: As many others have pointed out to you, you don't get access to the reasoning traces anymore on the subscription accounts. And the article also clearly states:
"Chinese developer communities assert [selling logs] is happening in at least some cases, but whether proxy operators are systematically harvesting and selling these logs, and to whom, remains unverified. However, downstream distillation data does exist on the open web. Several datasets of Claude Opus 4.6 reasoning outputs circulate on HuggingFace with no clear source for the outputs. Theoretically, one can clean and sell similar distilled datasets to other model developers in China."
The article also discusses selling logs for other (far worse!) purposes than for training, like blackmail.
So overall this article reads very, very different to your claims. Nothing in the article suggests or supports the idea of large-scale coordinated "distillation attacks". Instead it paints the picture of a naturally emerging grey-market response to access control blocks, consisting of many exchangeable individual actors: "Almost no one operates the full chain. Most participants own one or two links and monetise those well, resulting in a resilient, modular system."
Importantly: Nothing in any of this looks ethically worse to me than Meta using pirated books for training. And nothing suggests that OpenAI or Anthropic were more ethical than Meta when sourcing their material.
The situation strikes me as morally ambiguous. The resellers are:
1) selling Anthropic's products at a 95% discount and redirecting Anthropic's own customers to themselves. A customer is far less inclined to buy directly from Anthropic when a reseller is offering an identical product for 10x less. This situation is highly similar to internet piracy.
2) keeping the token logs from Anthropic's products and selling them to competitors, so those competitors can build their own equivalent models. The resellers get paid per token log they deliver. This situation is highly similar to espionage.
May I ask you a personal question? What is motivating you to take up the frontier labs' cause in this way? Not a rhetorical question.
For my part, I'll happily disclose that I have an axe to grind. I think the major AI labs are an aggressive form of a cancer that's been ravaging our society. I want to see them fail, of course -- but more than that, I want to see the public develop an immune response to this.
I just can't wrap my head around why someone would expend so much effort speaking up on their behalf. They have, after all, highly compensated PR people doing that for them!
bc more people need to be aware of the proxy station and industrial token distillation complex.
many people i've replied to refuse to believe this is going on.
once you realize what's actually happening, and that you can get Chinese-lab-subsidized tokens at a >95% discount, why would you ever pay full price for overpriced APIs?
if Ford bought hundreds of millions of dollars worth of Hyundais, put extra instrumentation in them, and resold them at a discount to customers who agreed to the instrumentation in exchange for the discount, is Ford doing industrial espionage?
1) Anthropic tokens via subscription aren't sold at a loss, they're sold at cost.
2) Subscription plans are not sold in hopes of eventually gaining a monopoly position. They act as a loss leader designed to get a foot-in-the-door and funnel companies into costly enterprise plans, where Anthropic can charge full API rates.
Why should other people be responsible to make their business model work? They hold like all the money, if they can't make it work please shut down. Companies like OpenAI have already broken the public trust by breaking their non profit promises. They don't deserve any politeness at this point.
Someone was claiming that it's not an issue for these proxy networks to create thousands of bot accounts and resell Claude's output because "they are buying the product" and "They pay for it!", which is a nonsensical position.
I responded that these resellers don't always acquire these accounts legitimately. They often use stolen credit cards, educational discounts, or resold compute credits to acquire them at essentially zero cost. They're not always paying customers.
That's one reason token resellers are able to price so cheaply, they acquire the goods for free.
> They often use stolen credit cards, educational discounts, or resold compute credits to acquire them at essentially zero cost. They're not always paying customers.
Yeah but where are you getting this from? I've seen this claim many places but only as pure speculation. No proof, just bold faced assertions.
The china talk page points to an article about crypto currency stolen credit cards. If I believed every random X account i would have to believe too many false things. Not really convinced my guy.
assuming the k3 model weights do indeed get published, if your model of the world is "achieving RSI is beneficial and K3 has done so," this feels structurally different from ordinary industrial espionage, because the knowledge has enriched the commons
more like silk than capacitors
if, again, your model is that RSI will be beneficial, why wouldn't making it available to all unlock more benefit globally than not doing that
Chinese resellers acquire hundreds of Claude Max 5x accounts and set up a custom proxy server. Customers point their ANTHROPIC_API_KEY at that proxy, and requests are routed to Anthropic through one of those hundreds of accounts. Because one $200 Claude Max 5x account gets the equivalent of ~$2000 in of API credits, these resellers can resell Anthropic tokens at a massive discount, undercutting official API prices by more than 90%.
To cut costs even further, these accounts are funded using educational discounts, startup credits, or stolen credit cards.
The resellers log all data traveling through their proxy networks, which they then resell to Chinese labs as high-quality training data for significant profit. https://x.com/xkajon/status/2050445443889525235
Why should anyone care? I couldn't give a single fuck, in fact if what you assert is true (definitely not proven), I applaud Moonshot - seems like a very smart way to operate.
The entire debate about distilled vs not distilled is just academic. As an end user, I don’t give a crap how the model was trained as long as it does the job affordably enough
I see all of AI as theft anyway so it makes no ontological difference if the theft was from a human or from another AI
If your business leaks to your customers… if you can ask your product for its inner secret sauce and it readily gives away the goose. That is not a moat.
The desire to accuse China of just copying is like 20 years out of date. It’s been wrong since some people on HN were in diapers.
People are going to be gobsmacked when, in our lifetime, China becomes a world power comparable to the U.S. Probably still poorer per capita, but at Spain/Italy levels, not third world country levels. And they’ll be shocked at the implications of that on the world economy, migration patterns, etc. There will be fields where China is a global leader, and Americans and Europeans will have to learn Chinese and move there, or else be stuck in some satellite office of a Chinese company. We’re all in Europe circa 1895 not realizing the behemoth America will become in WWI.
I am still shocked Spain/Italy and USA are considered 'first world' countries. We are not in 70s or even 90s anymore. I've been to China in 2011 also thinking I am visiting some huge village but...that was the most futuristic trip I ever had. I was surprised by the penetration level of the mobile devices - everything had a QR code, you could buy/sell/send money, pay services all with a single tap on a phone.
China has pursued a phased approach to growth. Instead of trying to pull everyone up at once, it’s trying to get top tier cities to a high level of development before moving to other cities. Shanghai already has a GDP per capita (PPP) that’s comparable to Madrid. But there are provincial capitals with almost 10 million people that are less than half or less than a quarter of that.
Visiting cities is also a misleading way to compare the U.S. in particular to anywhere else. I have family in town in Mississippi that has less than 10,000 people. But the town has a household income over 60% of the national household income. Cost of living adjusted, they’re about as well off as someone in a top tier city. Someone with a median household income can afford a newly renovated, 4 bedroom, 2,500 square foot house.
I've been in China in 2015 and like anywhere else in the world it was very mixed: some urban areas like central NY or central Madrid and Milan (or much shinier) and some rural areas like 200 year ago, but inevitably with electronics.
Basically in every country of the world you can travel one hour from big cities and get in a place deep in the fields or the woods with very different needs and dynamics from the city. They could be different countries and maybe both cities and countryside will be better off if we could have fractally composed states with different laws and regulations.
Americans optimize for hyper-individualistic convenience. The average one-way commute in Dallas, Texas is under 30 minutes. The average in Tokyo is 48-50 minutes. And the guy in Dallas goes to work in a perfectly climate controlled bubble where he doesn’t have to interact with anyone else, while the person in Tokyo is crammed into a rush hour subway car.
I love Tokyo too (never been to Shanghai), because I’m an asian collectivist at heart. But you can’t really compare across cultures when they’re optimizing for different things. Americans are very wealthy and spend a lot of their wealth optimizing to never have to be near other people.
Paris has a metro station everywhere at least in what a tourist can assume to be an enlarged city center. Tokyo is another city with a lot of metro stations. Manhattan too, at least up to Central Park (but 20+ since my last visit.) I don't remember Shanghai to stand out positively or negatively, but 11 years can be a long time.
I mean, you have those kind of luxuries even in the poorest of countries in Asia, it’s just that there’s still a huge discrepancy between rich and poor, city vs countryside.
It’s not difficult to find areas in all these countries that are significantly less developed than Spain/Portugal’s underdeveloped areas. It’s just not as black and white as you seem to suggest.
(I come from EU but have been living in various countries in Asia for over a decade)
> I am still shocked Spain/Italy and USA are considered 'first world' countries.
They're a mix. Rural southern Italy isn't the same as e.g. Milan or Venice. I've walked from 1st world to third world within a few blocks in San Francisco. It's a slightly longer walk in Cape Town.
> I was surprised by the penetration level of the mobile devices - everything had a QR code, you could buy/sell/send money,
I've has exact same experience in places in Africa (1). Yes there's poverty and crime, but also if the technology is affordable, effective and reduces the need to handle cash then it's adopted fast enough.
People's understanding of that part of the world is also decades out of date. Mobile devices actually "leapfrogged" the wired telecoms network rollout (2), but that was decades ago. Africa is huge and diverse, and it is not going to be China this decade, but also it's changing fast.
And it might be China-aligned as China positions to be a reliable trading partner with affordable goods. It's possible that affordable Chinese solar-battery electricity systems will cause another leapfrog. This includes Chinese EVs (3).
Some day China can pioneer in science or technology but the current claim about Chinese companies leading AI development is ridiculous given the evidence of distillation and the fact that like 95 percent of science that lead to the current state of AI happened in either North America or Europe.
To be honest if you want to list academic papers that lead to the current AI models the majority is either done by Google Research or sponsored by Google.
In 2017 maybe. This chart shows last year’s Neurips accepted papers by country and institution (top 50). What is missing here is that the papers from American institutions also have mostly Chinese authors. Europe is sliding and Singapore has more papers than Canada.
US is still winning because of their hardware dominance. Also they have astronomical budgets and much better financing. They throw money at an industry until they win. Whereas China throws lots of (educated) people at it. 38% of top AI researchers today have Chinese education and origin^. And hardware dominance will change in the upcoming years.
From the Hoover Institution’s analysis of the team behind DeepSeek:
“We find striking evidence that China has developed a robust pipeline of homegrown
talent. Nearly all of the researchers behind DeepSeek’s five papers were educated or
trained in China. More than half of them never left China for schooling or work,
demonstrating the country’s growing capacity to develop world-class AI talent through
an entirely domestic pipeline. And while nearly a quarter of DeepSeek researchers
gained some experience at US institutions during their careers, most returned to China,
creating a one-way knowledge transfer that benefits China’s AI ecosystem.”
That was from a year ago.
Consider that on top of this the country was starved of access to Nvidia chips - and therefore accelerated its development of Ascend chips, and it’s clear they are undeniably leaders in AI research and development. Not the only ones, but the achievements are crystal clear.
Exactly. China is a real tech power now, just like Japan and Taiwan. The U.S. is ahead in a lot of areas of technology, but China has home grown talent that is taking the lead in other areas. And unlike Japan and Taiwan, China has a much bigger pool to draw from.
Okay but I cannot stress this enough: no one cares.
It's international politics. The rules are optional, and written on the back of whoever agrees to enforce them.
If you're going to run around declaring AI is a strategic advantage vital to national security, then guess what? Stealing it is a great idea. That you stole it is only a problem if it means you're not developing the ability to support that work locally as well, and China seems to be doing very well at building it's local talent and support network.
If you ever listen to Russian propaganda, there's a similar theme: every big idea, everything good, all of it was definitely first developed in Russia - only Russians could ever have thought of it. Of course, Russia isn't actually a world leader in any of those things, or able to execute on them.
Which is what America is sounding like more and more these days.
> If you ever listen to Russian propaganda, there's a similar theme: every big idea, everything good, all of it was definitely first developed in Russia - only Russians could ever have thought of it. Of course, Russia isn't actually a world leader in any of those things, or able to execute on them.
When I was a kid watching Star Trek VI, I was confused by the line "You've not experienced Shakespeare until you've read him in the original Klingon".
And then I learned about how the Klingons (especially in that film) were a stand-in for the USSR.
But that was more of a jab at literary snobs who would tout "Homer in the original Greek" or "Marcus Aurelius in the original Latin" or "Old Testament in the Original Hebrew". It has been such a meme, probably for centuries. Because it was not so long ago when university students were actually conversant in many classical languages such as those.
This isn’t even controversial assuming you’re talking about the real world, economists freely admit that. It only holds for spherical markets in a vacuum.
Almost all markets depend on some form of regulation whether its as simple as "leave everyone alone but no stealing" or "every participant has to source every object through mountains of red tape."
Thus far the US has not really chosen to go the Chinese rare-earth method yet. The problem with distillation attacks is the end result is everyone who is not doing them is going to deal with some kind of regulation whether it's complete loss of access, or the amount of control you'll have to give up to access them will be ridiculous.
Sort of like the "stealing music is fine" but "lets freak out now that it's producing visual art", in the end the entire thing is a social construct. Whether this is treated as theft or "business as usual" is entirely societal.
Eventually the gap will close, unless there's a major breakthrough that hasn't been made yet.
Given these models could not have been trained in the first place if they had to license every line of random fan fiction on the internet, I think distillation also being fair game is a tradeoff everyone should be willing to take (unless they want to decelerate, but that's a different conversation).
We're still in the early days of the AI industry timeline(relative to traditional industries). Not everything has yet been litigated.
Taxes on AI subscriptions or AI capable hardware, to financially compensate IP holders for (potential) IP theft, could very well arrive in the near future, once the industry is mature.
If this shocks you and sounds preposterous, I'll remind you that in several EU countries, we still pay extra taxes on any and all storage mediums and on devices with built-in storage (tapes, CDs, DVDs, HDDs, SSDs, tablets, phones, etc) simply because they can be used to store pirated content, decisions based on laws from 50-100 years ago, and the money goes to the national unions and associations of music and arts IP holders. It's basically a lobby pushed and government legalized extortion racket that no voter agrees with or can change but has no choice but to conform either way.
So I guarantee you in the future, it will be the same for AI subscriptions and hardware capable of running LLMs locally. Every time you purchase a Claude or ChatGPT subscription, an Nvidia GPU, Intel/AMD SoC PC or an Apple/Qualcomm powered smartphone, you'll pay a government enforced tax to the likes of Sony, Axel Springer, etc. for licensing their IP, whether you want to or not. In the EU at least. US maybe not.
That is incorrect. Anthropic paid $1.5 billion in compensation to copyright holders for use of their content in training data. OpenAI pays hundreds of millions per year across 150+ licensing deals for access to copyrighted data. Meta and Alphabet have similar arrangements.
Under the settlement, Anthropic was forced to delete the pirated data they were training on.
Chinese labs can still train on pirated data. I doubt the Chinese models operate under similar licensing agreements.
Not at all. The ruling came from a federal district court, and since it was settled early, it was never reviewed by a higher court. It doesn't set a national precedent across the U.S.
There are more cases in the pipeline. The massive NYT vs OpenAI is still ongoing. Nothing will be "settled" until this makes its way to the Supreme Court or Congress steps in.
they didn't pay yet, because court challenged settlement as inadequate.
> I doubt the Chinese models operate under similar licensing agreements.
US corps likely pay licenses when afraid to be sued, or have troubles getting that data, otherwise they just take data, which was demonstrated many times. The same apply to Chinese corps, alibaba totally can be sued in US.
China is infamous for weakly enforcing copyright law. Even when it is completely obvious that Chinese labs are training models on pirated data, US copyright holders face a virtually impossible task of proving it in court. Those lawsuits won't go anywhere.
What are the most high-profile examples of the "tons" of lawsuits resulting in Chinese companies being banned from doing business in the U.S.? Isn’t it usually more action by the government - executive orders, etc?
I believe mechanics is following: US corp sues Chinese, asks for preliminary injunction to stop selling product for example if there is strong evidence some IP for example was stolen etc. Then they litigate, and settle somehow.
That 2024 article says "US sanctions" in the first sentence, but it's paywalled, but https://en.wikipedia.org/wiki/Hytera#United_States first mentions a 2019 US law that first partially banned them, with the US government subsequently expanding it to a general US ban. After the initial ban it appears Hytera was involved in a suit with Motorola and got a worldwide(!?) ban as a result of it in 2024, but the ban was lifted on appeal after 2 weeks (just after the SCMP article). So it appears Hytera was first banned by US law, then got a 2-week worldwide ban from a US suit. (I'm just relying on the linked sources and have no personal knowledge of all of this.)
>> There are tons of lawsuites which resulted in banning Chinese companies from doing business in US
> What are the most high-profile examples of the "tons" of lawsuits resulting in Chinese companies being banned from doing business in the U.S.? Isn’t it usually more action by the government - executive orders, etc?
In response to "What are the most high-profile examples of lawsuits resulting in Chinese companies being banned from doing business in the U.S.", the one example given was from 2 years ago of a ban that lasted for 2 weeks (separate from its 2019 onward government bans)?
However, if the claim is that companies (including Chinese) can face significant fines from IP lawsuits, I agree.
That's like saying someone is a big proponent of community law and order, and they donated $1000 to the county sheriff when actually they got caught drunk speeding in a school zone.
A false equivalence. A more correct example is: Anthropic was speeding, got caught by the county sheriff, and paid the fine. Anthropic stopped speeding.
Meanwhile, Chinese labs are speeding in a different county. Everyone knows they are speeding, yet the sheriff won't pull them over, so they just keep doing it.
This lax enforcement gives Chinese labs a structural advantage over American ones.
Destruction of Materials: In addition to the monetary compensation, Anthropic has agreed to destroy the two libraries that allegedly contain the pirated works, as well as any derivative copies originating from those sources. Anthropic must certify in writing to class counsel that the destruction has been completed and that the allegedly infringing materials are permanently removed from its systems.
The libraries in question were Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi).
If Anthropic is somehow training models on deleted data, I'd be quite impressed.
After the fact. They did the same thing Youtube, Uber and Airbnb did: Break the law, eventually get caught, cut some deal where they pay a pittance and keep doing the same thing but now with leverage on their side.
How is distillation an "attack" but gigascraping the Internet to the point of crashing servers and everyone needs Cloudflare and Anubis now not an "attack"?
I'm not aiming for a what about kickflip here: I'm saying we need to either agree on some rules or stop crying foul. Maybe the coherent legal theory is that neural networks and intellectual property don't interact. That would be weird but it would be consistent, a market could price it, I could do coding stuff and know if I was illegaling.
But this weird gerrymander that no judge will really rule on in an emphatic way is like, bad for the planet, bad for markets, bad business.
There are a lot of reasons to look forward to DeepSeek Huggingface drop kicking the unambiguous frontier weights in like, November, but I think my favorite one will be "who's distilling now bitch?"
I think you've basically got the legal theory. Training a neural network isn't prohibited by copyright law so if you can legally get your hands on something (e.g. by sending a GET request to someone with rights to serve the contents of their web page, or by buying a book) without signing a contract to not train on it, you can train on it.
But the American AI companies only let you query their models if you first sign a contract to not train on the output.
It's hypocrisy and unfair, but I think there's a strong legal argument for it.
Of course China can simply decline to assist in enforcing that contract... But I would expect US courts to do their best to.
Now THAT'S doing some heavy lifting lmao. The vast, vast, VAST majority of the original datasets were from pirated books and the like. Also, arguably a robots.txt is the exact mechanism to follow to do the mass GET-ing, yet the AI cos choose time and time and time again to simply ignore it and be as abusive as they possibly fucking can
Yes, anthropic and openai have really been brought to their knees and ipos cancelled because of the legal consequences of obtaining their training data.
This would have been a problem but it turns out that Anthropic is actually valued multiple orders of magnitude more than a copy of all the books in the world. So they survived the significant legal consequences.
In a just world, the punishment can be more than just the sum of the direct damages, otherwise there's no incentive to stop reoffending.
Anthropic (and friends) proved they're willing to do obviously illegal things, and it didn't end them. Why do we think they stopped after doing it once?
I mean the latter, but more narrowly: China would never allow the United States to have a monopoly on machine intelligence if the only thing standing in the way of a domestic alternative was the Anthropic ToS. In general, I think that China is willing to agree on certain things relating to intellectual property. But not on this, it’s too big.
The US is already publicizing the way they are using Claude with Palantir for war gaming purposes. It’s a matter of national defense. Contract law has no meaning here.
> but I think there's a strong legal argument for it.
Maybe today. I doubt it tomorrow. Legal and not legal, largely, has to answer to the population sooner or later. Ultimately, humanity decides legality. And I don't think the frontier labs will get a pass from humanity in the midterm, let alone the long term. I think you'll see the rules change towards something more "intent" driven. And there's absolutely no difference in intent between Frontier labs and everyone chasing them.
Frontier labs just want the door closed behind them, as do their investors, because they know the money will never be recouped if others can do the same magic tricks.
Eh, I think you've done a pretty good job summarizing a collection of settlements with a few narrow bench rulings for seasoning. I'm not sure I follow you to it being a coherent legal theory. Buying a book in a bookstore is sure legal, and excerpting from it for e.g. literary criticism is pretty settled. Downloading every torrent of all e-books ever is pretty clearly illegal (or at least it fuckin would be if I did it). Pretty sure like, multiple labs have been popped for that though.
Situation right now seems more like a fragile detente: if you got a Hill staffer drunk and hounded him long enough he'd probably be like "God damnit the market will fucking tank if we don't get these two IPOs out north of a trillion. And don't even get me started on how I'm going to sell Chinese AI to a Senate that still calls people Nipponesians when no one is looking. We're doing the best we can alright, get off my back man."
We have a situation, but it's not exactly A&M Records, Inc. v. Napster.
> Downloading every torrent of all e-books ever is pretty clearly illegal (or at least it fuckin would be if I did it). Pretty sure like, multiple labs have been popped for that though.
Oh it is, and at least anthropic has paid $1.5 billion and deleted there torrented copies and not released any models derived from them as a consequence.
The thing is it turns out to be not that expensive to just buy a copy of every book legally and scan them. And there's even precedent that this is legal predating LLMs (Google books)
Great, let's go down to the courthouse and get some sworn testimony as to the ownership, value, condition, and so on and so forth of the bridge. And some document review and discovery run through professional legal firms under the same conditions. And perfectly reasonable and verifiable explanations as to why you own the bridge and are selling it (namely that you bought a copy of literally every book in existence in the meantime).
Facts are in fact knowable, and the US legal system is in fact not terrible at getting to them.
I think you're right to point out that historically the rule of law in the United States has been very robust by the standards of whatever era, it's been a tremendous advantage in attracting business and capital and talent, it's good stuff.
But we've gone through some pretty weird times too. Turn of the last century was pretty tech billionaire edits, reconstruction was uh, not smooth, it's a mixed bag.
And most takes I hear seem to acknowledge that this is one of those weirder times: serious election fraud rhetoric from most everybody from 2016 to the present, very politicized courts (on both sides to be clear), very soft on anti-trust, very soft on adventurous accounting. The Epstein files and like, no consequences (pretty much uniquely for a developed nation with Epstein people). It's weird right now.
And I think I would be hard pressed to think of a weirder part of this weird time than the rule of law meets AI. We can haggle on where laws end and norms begin (stare decis being maybe the midpoint), but in the 90s, the Justice Department got their brass knuckles on for a lot less.
I don't think it's a simple "the law works nothing to see here" story.
I broadly agree with your take on the state of the US - but this is a case where given the specific facts at hand I'm confident it still got to the truth.
I can understand why as someone who didn't follow it and the more corrupt legal developments closely you wouldn't be confident in that.
Knowlege should not have ownership. Training and distillation should be allowed
Granting people some form of control over knowledge only serves the public interest inasmuch it provides incentive to create more of it. Mass media, effortless duplication, and copyright extensions had already broken this to the point where control of knowledge was suppressing creation of new knowledge more than it facilitated.
The world has changed, we need a mechanism that works for the public interest that applies to the facts as they now are.
I think it's worth stepping back here and pointing out the obvious. Y'all waging war on math. And I'm sorry, but that's the computing equivalent of legislating gravity.
Apologies for repeating myself here, but what you call "distillation" is function approximation.
I feel for the teams at Anthropic and Open AI, but unlike startups from prior eras; Anthropic and OpenAI have decided to be in the business of selling compute. Not creating a product that uses compute, but a product that's math running on compute. This is different from what Google is (or, rather was. As always, RIP Google 1998-2019).
Google's algorithm might be math, but Google search isn't. Google search is a process that's continuously operating in the background. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Now, let's compare that to AI models. When Anthropic serves Mythos / Opus etc, they're taking input or x from their user, doing compute, and then serving the result of the Mythos / Opus function, i.e.,
According to Stone-Weierstrass, given enough values of y for f(x), anyone can approximate this function.
The fidelity and sophistication of this approximation definitely requires a lot of cleverness and effort, and it is arguably an imposition on Anthropic and OpenAI. But on a long-enough timeline, they don't even have to poll Anthropic or OpenAI. As the internet is flooded by PRs, content, emails written by Mythos / Claude, and just people otherwise sharing the results of Claude prompts, then there's an ever increasing set of data to approximate the f(x) that's f_Claude.
Eventually, in the future, anyone will be able to create a good enough approximation of the f_Mythos. Which is Anthropic's product.
Anthropic and OpenAI can now wage war on mathematics and the open-ended compute. Or, they can adapt and build a better product.
Choosing Option B was the Silicon Valley option / choice. I think the OG large-scale Valley lobbying effort, the Semiconductor Industry Association, was unique in that it prioritized and chose to do real research.
I like your point that there is so much content being created by LLMs that at some point there’s enough to perform something like distillation without even needing to interact with the LLMs directly.
Look how hard Anthropic is to even be able scroll back on your conversation, or look at the thinking tokens or subagents. They want to keep everyone coming back to the watering hole but never to learn how to dig a well.
I wonder, how does distillation deal with unprobed spaces in the knowledge landscape? Is a distilled model worse in some niche area that was not probed? Presumably, this is why frontier labs dont distill their own models internally to release them to the public as a servicable frontier model.
if distillation is the key, why the fuck all other competitors do not release competitive models? and only Chinese can distill this great?! Am I smoking too much?
are there any "open source" efforts to do distillation? Like some place one can submit one's anonymized chat logs? So they can be pooled and used as an open training set (similar to OpenCrawl)
Pulling on this thread, if the model companies become commoditized and make no money then who is buying the hardware? Seems like it would be the next shoe to drop
I suspect that distillation attacks may be slightly exaggerated.
Most of the training data used during fine-tuning is now synthetic data. You can't just repeat the same stuff twice, therefore another LLM is writing a text book that is explaining a topic in detail, ideally without any gaps in the material.
The fact that API based distillation is even a conversation right now makes me feel like the U.S. has their heads so far in the sand that it’s not really excusable.
These Chinese labs are producing novel models, publishing their techniques and sharing their open weights and the first topic of conversation is how they stole from U.S. AI labs.
Setting aside the fact that it doesn’t make any feasible sense to do API distillation, these models are outperforming frontier models on a number of benchmarks, and often times run more efficiently by several orders of magnitude.
We have to stop crying distillation, it’s getting embarrassing and at this point feels even a bit delusional.
> We have to stop crying distillation, it’s getting embarrassing and at this point feels even a bit delusional.
It's a PR campaign - when they say its an "attack" they don't mean on Anthropic - but on America itself. What kind of American can let such a brazen attack go unanswered? At the very least, they ought to demand the dangerous, pinko, stolen models be banned in all 50 states, and pay whatever price demanded by the patriotic, freedom-loving, all-American AI labs that can never be accused of stealing.
It’s so funny to me that Anthropic can make claims like this one with zero evidence provided.
DeepSeek and others like Minimax are publishing deep research on Multi-Head Latent Attention and Mixture of Experts, Multi-Token Prediction, novel Sparse Attention approaches, I mean they trained long context models on a fraction of the resources and gave everyone the recipe.
Chinese labs might not have the funding of labs like Anthropic, but at least they provide the receipts.
> This behavior is exactly what you'd expect from a model distilled from Claude.
This is not at all what I would expect because it's trivial to change the training data to replace Claude with Kimi. In fact I'd argue it's almost certainly not saying that due to distillation.
I encourage you to review the links before committing to a position. The writeup on K3's anomalous trans-model identity is very comprehensive.
K3 reproduces Claude's internal model identifier when prompted, something which the real Claude models themselves do not emit. This is highly suggestive that K3 was trained on Claude metadata (API logs, tagged synthetic data), rather than Claude's chat outputs.
>This is not at all what I would expect because it's trivial to change the training data to replace Claude with Kimi.
Wait what? The reason you wouldn't expect it is because if it was distilled, it would be easy to get rid of self identification? Is that any less true of a non distilled model? I suppose there's lots of ways to interpret it, but the idea that self-identifying as Claude is affirmative evidence that it's not distilled seems to get the weight of the inference exactly backwards.
By evidence I mean logs, I mean IP addresses, I mean timestamps. They claim millions of requests, let’s see literally any of them?
I don’t consider a tweet by Denise Wu, who works at Anthropic, to be reproducible evidence.
I don’t consider “Caveat: fully AI-generated research.” To be someone taking time to analyze anything in great detail.
Because two AI models produce vaguely similar front-end styles when generating similar prompts I also do not consider to be of much value?
I think this is what I mean when I say the U.S. has its head in the sand. The Chinese labs are releasing ~60 page research reports with citations and analyses and evidence and Anthropic is throwing up defensive blog posts with zilch. I’ve seen more detail in a tech blog from Uber than anything I’ve seen from Anthropic.
"Zero evidence" as you claimed earlier isn't accurate. You've moved the goalposts from "evidence" to "raw internal logs I can independently audit," which is a different and very high standard. Sure Anthropic didn't publish logs, IP addresses, timestamps, or account IDs of the accounts involved. But that's true of any cybersecurity breach/abuse disclosure ever made. Companies are furtive to reveal how they detect fraud, because doing so exposes the signals used to detect bad actors, and makes future abuse easier. Not revealing the "evidence" you're asking for is industry standard practice. You're complaining that Anthropic is following industry standard practice, and conveniently defining the "evidence" you need as something Anthropic is never going to publish.
> I don’t consider a tweet by Denise Wu, who works at Anthropic, to be reproducible evidence.
Is the issue here that she works at Anthropic? Because Denise Wu doesn't work there.
> I don’t consider “Caveat: fully AI-generated research” to be someone taking time to analyze anything in great detail.
The experiments were run by Ryan Greenblatt, who is a real AI safety researcher (at Redwood Research).
The identity experiments and Greenblatt analysis are trivially reproducible. The methodology, code, and metrics are all there in the Github repository. You can ask your preferred AI to independently replicate these results, and it will give you a result within an hour.
You’ve also reduced the evidence to “two models producing vaguely similar front-end styles,” which is not what either analysis shows.
From the analysis, Kimi K3 identifies itself as Claude 15% of the time. How do you explain that? Qwen and GPT identify themselves as Claude 0% of the time.
If a long document is too much analysis for you, someone else made a simple chart which measures the KL divergence between Kimi K3 and other major models. They found K3 is unusually similar to Fable 5 & Opus models. That is, Kimi K3 has an very similar style and phrasing to that of Anthropic models. That behavior is expected from a model distilled from Claude.
"From the analysis, Kimi K3 identifies itself as Claude 15% of the time. How do you explain that? Qwen and GPT identify themselves as Claude 0% of the time."
Qwen and GPT have special guards that trigger when asked to identify, Kimi doesnt. I dont understand the argument. Kimi is an LLM and does not know what it is. It will give you the most likely answer which sometimes is Claude.
I wouldn’t say I’ve backtracked- I think I’ve been incredibly consistent here. Chinese labs are releasing open weight models, research and analysis. Anthropic is not. They haven’t produced any actual evidence of distillation themselves and what they have presented is tenuous at best.
While it sounds like a lot, do you suppose 3.4 million sessions come even close to being sufficient to train a frontier model?
Assuming each session was 10,000 words each, that's 34 billion words; lets call it 50 billion tokens (0.05 trillion) unfairly pilfered from Claude. That left Moonshot needing to scrounge for the other 14.950 trillion training tokens required for a baseline frontier model.
Distillation attacks aren't about replacing the entire pretraining dataset with questionably sourced synthetics. It's all about post-training.
Train your own base model - but tune it off Claude output to make it perform more in line with Claude. Yoink the products of Anthropic's expensive SFT, RLHF and RLVR work for yourself by training on the outcomes.
The post-training datasets are small, but they are what controls the final model behavior.
> Train your own base model - but tune it off Claude output to make it perform more in line with Claude
Is that actually genuine distillation though? Distillation suggests the core model is being pre-trained using output from another model. For the above to work, you have to already have all the core intelligence trained into your base model.
If distillation just comes down to post-training then it's tantamount to admitting that the Chinese base models are just as good as frontier US lab models. Because you can't post-train frontier intelligence into a model. It has to be there in the base. Then you can change how that intelligence is expressed through post-training.
What's in the base model is "bits and pieces of intelligence".
You have to bring those bits and pieces together, put them into the right shapes and fill in the gaps to get a model that actually performs. This is what post-training is all about. It's not at all a trivial thing.
Reasoning, tool use, agentic behavior - all of those are post-training performance gains. Getting a good well trained base model is putting your foot in the door of frontier performance - post-training is how you actually get inside.
See: GPT-4.5 vs o1. One went for "build a bigger better more capable base model", the other went for "take the old base and post-train it for advanced capabilities". The results: a wider base with basic post-training loses to a narrower base with advanced post-training. Or, hell: GPT-3 vs GPT-3.5. One was largely a research lab curio, and the other kicked off the AI revolution as we know it.
The gains compound. Getting a better base model with the same type of post-training helps, see: the jump from Opus to Mythos/Fable. But post-training techniques account for a lot of the performance juice.
And yes, reasoning trace post-training distillation is "genuine distillation". As is logit distillation in pre-training. "Distillation" isn't a single training recipe that you have to follow to a tee - it's a large group of training methods. I've seen plenty of wacky things like inverse distillation bootstrap and post-training self-distillation that use distillation in strange ways at different stages of the training run to get results.
How does yoinking outputs from from prior generation Claude model and post raining on them result in a model competitive with the latest generation? That doesn't add up - nevermind Anthropic hasbeen summarizing thinking tokens since January to counter distillation.
Do I really have to explain the shape of AI training pipelines to you?
Train a big, wide base model with a lot of potential. Mid-train or post-train that on Claude Opus 4.5 reasoning/agentic traces (i.e. Claude Code data from Chinese API resellers) to make your model approximate a high baseline of chatbot behavior, reasoning, agentic work and tool use.
Then run your own expensive SFT, RLHF and RLVR on top of that yoinked baseline to dial it in further.
Actually doing RLHF and RLVR is extremely expensive. Distillation gives you a lot of dense, high quality post-training signal for cheap. This can get your model into the basin of "the right way to tackle this kind of problem" without a frontier lab compute budget. It's a big shortcut that gets you closer to the target - you can take it from there and build on top of it with your own work.
Also, it's unclear whether "summarizing thinking tokens" actually ruins distillation, or just makes it work worse. I'd bet on the latter, really. Because it's an approximation game, and summarized reasoning is still a better approximation of true reasoning than most of what you get online and in pre-training datasets.
3.4 million is the number of sessions Anthropic detected. The actual number of Claude sessions trained on is likely >100 million. There are tens of thousands of accounts funneling Claude sessions into Chinese labs https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
They are used for post-training, i.e. calibrating the model to understand and use tools/command line more effectively.
> 3.4 million is the number of sessions Anthropic detected. The actual number of Claude sessions trained on is likely >100 million.
That's an increase of only a single order of magnitude, increasing my estimate of exfiltrated tokens from 0.05 to 0.15 trillion - a far cry from the 15 trillion required.
> They are used for post-training
Possibly - it may be too much data for post-training, unless further curation was done. However, this is not distillation; you know it, I know it, Dario knows it, but "Distillation Attack" is a short, memorable, sciencey-sounding, political sound-bite with enough malevolence to be deployed on the floors of congress, or by the usual fear-mongering newstainment talking heads.
You're conflating pre-training data volume with post-training data volume.
Nobody is suggesting Moonshot used 15 trillion tokens of Claude data to pre-train a base model from scratch. That would be impossible and nonsensical.
This is entirely about distillation, which happens during post-training (alignment and SFT). Here, datasets are measured in millions or billions of tokens, not trillions. 50 billion Claude tokens is far, far than enough to copy Claude's reasoning logic, writing style, and tool-use ability to the pre-trained base model.
> However, this is not distillation
I don't understand how you're so caught up on the term "distillation". Distillation is using a larger model's outputs to train a (weaker) student model. Which is exactly what's happening. It's a standardized term that has been in use for a decade.
There is a lot of supposition going on your part and mine. IMO, Chinese labs are not dependent on OpenAI/Anthropic outputs; they definitely use the outputs, but along other training/post-training data.
Now that Anthropic hides the real thinking tokens in a way that precludes future CoT distillation, we'll find out which side is correct based on whether Chinese AI labs close the gap or not.
My bet is they'll close the gap; nothing about frontier AI is magic, once something is shown to be possible, experienced practitioners almost always figure out how to accomplish the same feat, though not always on the same way. This is why frontier US labs keep leapfrogging each other every few months.
This line of thinking makes no sense because it assumes that labs that distill from frontier models are doing nothing else. It's the classic "the Chinese can only copy" mentality, and it's going to end poorly for American companies.
I'm pretty sure that all labs are distilling each others' LLMs, maybe apart from Anthropic and OpenAI. It would be stupid not to do it, because it's cheap and effective. But that's not the only thing they're doing. If you think K3 and GLM-5.2 got this good only from distilling frontier models, you're not paying attention to Chinese labs' publications.
i never assumed that, and i do keep up with the publications. i'm also not saying it's a dumb thing to do! what i am saying is that empirically, it appears that distillation of a more advanced model is a required first step for them to train a borderline competitive, cheaper model. in effect, their training is subsidized by the frontier labs.
if this were not the case, then we would be observing chinese models that far surpass frontier models in capabilities, rather than "almost as good, but much cheaper", and we would be having a very different conversation. what happens to these efforts when the subsidy is cut off?
> empirically, it appears that distillation of a more advanced model is a required first step
I see no evidence for that.
> if this were not the case, then we would be observing chinese models that far surpass frontier models
It's pretty clear that the primary reason for the difference is budget and compute availability. Chinese labs have at least an order of magnitude less money than Anthropic and OpenAI.
> what happens to these efforts when the subsidy is cut off?
They will continue making progress as they do now, minus the benefits of distillation.
Agentic reasoning and tool use
Coding and data analysis
Computer-use agent development
Computer vision
Moonshot (Kimi models) employed hundreds of fraudulent accounts spanning multiple access pathways. Varied account types made the campaign harder to detect as a coordinated operation. We attributed the campaign through request metadata, which matched the public profiles of senior Moonshot staff. In a later phase, Moonshot used a more targeted approach, attempting to extract and reconstruct Claude’s reasoning traces.
I'm assuming you posted that as evidence for the claim that "empirically, it appears that distillation of a more advanced model is a required first step", but I don't think it is. It's just evidence that Moonshot distills Anthropic's models, which, yes, they do.
it is not a required first step for training a model, sure. but that's not what i claimed. what i claimed is that is how they are so significantly _reducing the cost_ of training one! how else do you think they are doing it?
Distillation from a teacher model solves the self-start problem, that is, building a model to the point where it reason coherently. Without distillation, solving self-start is incredibly difficult since it requires millions of high quality training samples. Creating that kind of dataset takes an enormous amount of effort.
Once a model becomes competent enough to perform complex reasoning, a teacher model is no longer necessary. The model can now reason about its own behavior and build a better version of itself through recursive self-improvement (RSI).
the question was: what is the endgame for the stated "second class labs" strategy of distilling their frontier competitors then undercutting them on price?
Yes yes, we all understand the game-theoretic race-to-the-bottom you're describing here. Somehow despite linux being FOSS it still powers most of the important computing in the world. Can you explain how that works despite it being free? Once you understand that case I think you'll understand the game-theory behind how large projects can exist in the absence of traditional IP protection.
the obvious difference is the massive scale of data and compute required to develop and evolve these models, and the costs they impose on those building them.
Smaller budgets, slower improvement, less risk. They're not entitled to profits if that business model isn't sustainable. They're not entitled to a change in IP laws to protect their business model. They're not entitled to growing that fast.
Well, there is precedence: Google can scrape the web, but you can't scrape Google. Laws around compiled databases exist for a reason: you can't just copy the phone book if effort has gone into compiling it, it is itself copyrightable
That varies by jurisdiction. In the United States, copying the phone book (or otherwise copying facts from someone else's collection) has been legal since 1991:
Say it louder for the people in the back. All these complaints about "distillation" from frontier labs are bordering on felony contempt of business model at this point. It's great for us. Maybe it's bad for them but nobody other than shareholders really cares.
The optimal outcome for humanity is for oligarchs to spend trillions training a godlike AI, only for the precious weights to just leak. No "distillation" required.
The hand wringing over whether internationally located AI labs are "stealing" output from American ones is the funniest thing in a while.
It's international politics with people talking about AI success as a matter of national strategic advantage and survival. So at best "this was built off our work" mostly tells you that apparently you've got months of advantage when a new model drops before it can be cloned. That's certainly some sort of advantage, sure hope it represents a consistent ability to stay ahead and causes people to redouble their efforts.
Or...of course none of these companies are worth what they say, but the advantage is also not really that great, and a whole lot of people are just really worried about their stock payouts.
Thanks for the models guys, sorry for your losses. Once this reality becomes mainstream and undeniable, surely the bubble pops and then what then. Future model development stops? Becomes private? Becomes a public effort?
The existing models are still going to exist. As hardware improves, there will be a day where it might cost a tenth of a penny to churn through 100M tokens a second of Opus 4.8. Established compute providers will invest in improving the models incrementally when margins drive them to look there.
If "distillation attacks" happen, we have to conclude there is some value add in what model labs do. Regardless of how we feel about using existing human knowledge in the way they currently do, it's simply impractical to infer that everything that happens downstream of LLMs can not be an attack on some IP because of it.
So both things can be true: a) People infringe on Anthropics IP and b) what Anthropic did to build their models is legally questionable (or might be ruled illegal, even though I doubt it).
Authors do not infringe on IP when they read another's book, nor should the lumber company be able to dictate how I use planks and if I can resell them if i'm done with them.
You're framing it as if the added value of the author or lumber company, awards them consideration when somebody uses the products to create more value.
IP law was always a big mess, and these questions cross far into ideology instead of law; but I do not understand people who think we need an ideology where more IP-law is good for society.
It's more simple: They infringe on the IP by way of violating the ToS. If you violate ToS and the company suffers financial harm, they usually can (usually) sue you in civil court for damages.
There are some quite interesting legal implications here. If Anthropic has IP over output produced by agents, do they somehow have legal rights to code and documents produced by such agents?
Unless someone literally stole the weights somehow (which is not out of the question, I doubt either oAI/Anthropic have the capabilities to prevent a state-level actor getting those weights), distillation from generations is not infringement on anyone's IP nor is it stealing nor is it an attack. It can't be. As long as you pay for tokens you get to do whatever you want with them. Someone saying you can't doesn't mean it's an attack or their IP or whatever. They either sell the tokens or not. They can decide to not sell them to anyone, but again that's not stealing.
And their ToS are a joke. Imagine how people would react if MS had ToS saying that you can't use MS software to develop solutions that compete with MS. They'd be laughed out of the room. Somehow it's ok for token sellers to decide what you do with the tokens? Why? If you pay for something you get to do whatever you want with that output. Train, distill, whatever.
Its definitely an attack. Thats established from anthropics perspective. No one has a right to use Anthropic’s services in ways that directly violate the ToS and user agreements.
I guess I can see that, if you mean the targeted effort of creating many accounts w/ the intent of doing it at scale. Sure, they may see that as an attack. But again, it's only an attack from their perspective if you agree that using generations to distill is "wrong". I just don't see it, in general. You can't both sell tokens and decide that distilling is somehow illegal. Something, something, cake and eat it.
> Its definitely an attack. Thats established from anthropics perspective.
How do things get "established" from someone's perspective, exactly?
By that logic it is established from my perspective that Anthropic has no right to train on anything I've written that is publicly available on the internet.
Of course, they don't care about my perspective, but then again I don't care about theirs.
Anthropic’s model outputs contain no IP. This is actually a simple legal proposition (rare in this field!) that derives from the fact that only specific classes of IP exist: copyrights, patents, trade secrets, and trademarks. Examining each, it is clear that API outputs do not qualify. Anthropic disclaims copyright in outputs; the outputs are not patented; the outputs are not secret (a prerequisite to having trade secrets); and trademarks are irrelevant in concept.
The output of Anthropic's models is not Anthropic's IP, as that would destroy their market, if Anthropic owned all the software it generated, and all the content. So distillation, which is just using those outputs is always going to exist.
I'm pretty sure that LLM output is not intellectual property. Nobody owns it, and it can't even be copyrighted. So using output from Anthropic's LLMs in ways Anthropic does not condone is not IP infringement.
Considering they were the original infringers, I don't know how anyone can expect tears to be shed here. The best we can hope for is for all these cancerous - and they really are the definition of a cancer - money burning entities to all fall apart to distillation attacks like these.
Regardless of whether it’s intellectual property or it isn’t intellectual property, it doesn’t actually matter. If AI doesn’t stop seeing diminishing returns in scaling up, and it hasn’t yet in the 10 years since the attention/transformers paper, the advent of AI will be the most important development in the history of humanity. Controlling that machine, or at least having one of your own, is an existential problem for nation states. It’s like a matter of national defense.
Do you really think intellectual property laws will prevent this in practice? It’s like as if we said, “hey, USSR, you can’t make a nuke, too! We patented that already.”
Asking China to not distill our models down is equally as ridiculous.
I do like that you mention diminishing returns, because we are hitting them in building out all the external requirements for competing at the frontier. Even if model performance scales linearly with energy input, the top labs are now competing with other uses for that energy.
How far are we willing to go as a nation (and as a species) to prove out the scaling laws? Are we willing to sacrifice our industrial base? Would we rather train models or smelt aluminum?
It's unlikely the USA would be granted an exclusive patent for the atomic bomb given the well-established existence of prior-art in the form of nuclear fission on the sun.
(I actually appreciated your analogy, despite my lark)
Anthropic’s IP is basically null and void for how they created it. And they might not want to try and challenge this in court, considering how they had to settle for using text books they had no right to use
Answer the question "how much does 5 cents of LLM computation in July 2026 cost in July 2005" and you'll have the answer to your question.
Don't forget to account for all the costs. It's not just that CPUs are X times slower. Memory is X times smaller, too, and networks are X time slower. And all this hardware is many times more expensive.
If I'm getting my mental estimation right, training a 2026-frontier-class LLM in 2005 would be somewhere on the order of all the computation power in the world at the time. It's not that many more factors of magnitude before you end up at "all the computation power in the world up to that point".
Is this some form of rage bait? 2005 we hadn't the GPUs, we have today. There are other factors, but I think this is the big one. The mathematics of building an LLM are really old, we just hadn't the hardware to do the needed calculations.
Right. Therefor it's not simply a derivative of information. The hardware is required to build the model. Software as well. The model uses information, it is not "distilled" from it.
"Distillation" literally means to separate and take some components out of something. You can distill how a model works from a model. You cant distill a model from information because the information does not contain the model.
People are happy to conflate distilling with building because they dont like how the information was used. You distill how the model works from the model, and you build a model with information. Both could be morally good or bad but its not the same thing.
Not really, what the information actually is, matters a great deal. It's harder to get good results going from "nothing > model+weights" than "nothing + traces from known good sessions of other good model > model+weights", this is what the "distillation" part is referring to. If "information is information", you wouldn't even need to separate good from bad sessions while doing the training, which leads to somewhat obvious results if you don't.
Can you be more specific? I have no idea what you are trying to say.
To succinctly restate my point, you cannot distill a model from information because the model is not contained within that information. You can distill a model from another model.
Their point is that "training" and "distillation" are essentially the same. The difference between the words is whether the source material is output from another model, vs being some original text.
That argument is moot as distillation also requires a lot of hardware and software, if copying models was as easy as that, we would have hundreds of competing models.
We didn't have the compute required (GPUs powerful enough to parallelize forward and backward pass). This compute is what allows us to train from human knowledge or distillation.
because you had neither the chips or the information in 2005. You have probably on the order of 5000x to 10000x more GPU compute today than you had in 2005 and three to four magnitudes more openly available data.
The first "L" in LLM does the work. In 2005 you had no Github, Stackoverflow, Youtube, common crawl and no archive of digital ebooks.
> There was never any plausible explanation for why this wouldn’t happen.
What a nice post hoc revision of history. Distillation is still an active area of research, that you can distill models as easily as you can it genuinely interesting and absolutely not something that was taken for granted even 12 months ago.
Even 6 months ago this idea that 'using model outputs as training examples' was listed as the reason that all models would fail in the near future due to some spooky circular training catastrophe.
I think you’re being overly combative. It’s intuitively quite obvious that it’s incredibly easy to implement and the circular training catastrophe was only ever a conjecture. It’s kind of like releasing a crypto primitive without knowing a proof. Like… maybe it works, but you can’t assume that just because you don’t know how to break it. You have to remember that 100s of billions of enterprise valuation rely on frontier models being moats. The burden of proof is on those raising valuations assuming they will capture the full market.
I agree that hindsight is doing work here, but DeepSeek R1 from Jan 2025 seemed to heavily leverage distillation, and 18 months is an eternity in this climate.
This was always where this was heading, but we got here much faster than expected.
Once western governments declare it to be a "national security" risk for citizens to have access to open-weight frontier models, and once they classify using these models as acts of terrorism, what will that world be like?
Will using Kimi K3 come to be like how napster was in the olden days? Everybody knew it was technically illegal, but come on -- any track at your fingertips? But surveillance is quite more evolved now.
Or it will be like cannabis, where a guy in the neighborhood will low key rent you metered access to the 8x5090 rig in his basement he cobbled together from parts on ebay? Or everyone will flock to VPNs?
Or will the oppressors actually succeed? The same way that napster is long gone, and everyone accepts that they must pay spotify for a homogenized collection, where artists must take only a minuscule cut (more than napster though)... We'll be stuck with nerfed Cohere or Mistral models for open-weight options, as if they need more lobotomizing. Or else we can pay through the nose for Anthropic/OpenAI for "American Frontier" models which will fall increasingly far behind China.
Or else, like how Kindle Fire was subsidized by ads, we'll have "Kindle AI" where influence is sold to the highest bidder, where the LLM will tell us that smoking is actually healthy if big tobacco can engineer its renaissance by turning its lobbying dollars to pay-to-play, pumping its propaganda into the training pipeline for Amazon's extra commercialized line of ultra budget LLMs.
Basically a new iron curtain didving the world into digiatl blocks.The era of open internet/science is on its last legs with the potential forr bifurcation into incompatible ecosystems high , the onger the exchange is disrupted.
As recently as this month the USgov has donce a Wolf Amendment style declaration for the Scientific collaboration NSF while shifting its purview under the military.
To add to that its trying to rope as many countries into its Pax Silica idea intentionally to exclude China while simultaneosly coercing its 'allies' into using its nerfed offerings [1]
So maybe some isolated switzrland/singapore type locales would exist for US/EUusers to be able to dip their toes across the curtain legally without reprucursions.
> The Wolf Amendment is a law passed by the United States Congress in 2011, named after Representative Frank Wolf, that prohibits the National Aeronautics and Space Administration (NASA) from using government funds to engage in direct, bilateral cooperation with the Chinese government and China-affiliated organizations from its activities without explicit authorization from the Federal Bureau of Investigation and Congress
At this point, the United States will lose that battle most Countries in the world are going end up using electronics from Asia, that ship has sailed Japan, China, Korea, Singapore, Taiwan, Vietnam, dominate that area, China already dominates EVs, Drones and many other electronic devices, and with the way Donald Trump has picked fights, Europe, Canada, Australia, New Zealand, Mexico and many others are looking for other business partners.
If you need infrastructure done, China is dominating that area too. Rail, High-speed rail, Nuclear reactors, (near future Thorium reactors), Dams, Highway roads, bridges, Ocean ports, airports you name it, and they can roll it out, Transport ships, And if they don’t do it, Japan, Korea, Vietnam, and Taiwan do.
Is it too late? No, not necessarily, but America needs a regime change…
If rural America is that unappetizing, you understand you can just go live somewhere else, right? There is a very deep-seated hatred here that I suspect has little to do with actual "rural people".
America is what it is. The only thing that will change it is leaning in not bemoaning "rural people" on Hacker News.
If they outlaw open source models that'll just handicap American companies, the rest of the world will be running open source and have an arbitrage against US companies.
"Why would other countries, that don't share the same anxiety about China as the US, would be troubled with the this?"
It's the other way around.
There is a high likelihood that many countries of the "west" (the "global north"?) will outlaw, restrict, or otherwise control LLMs and the tools that enable them.
The US, however, is blessed with the first amendment which makes it extremely difficult to restrain speech in any form - including code.
> There is a high likelihood that many countries of the "west" (the "global north"?) will outlaw, restrict, or otherwise control LLMs and the tools that enable them.
US pressure is worth a lot less than it used to be; that's why other developed countries are urgently prioritizing digital sovereignty after years of technological sclerosis where they were happy to run on US-managed cloud infrastructure.
It's not just the tariffs and imperialist/autocratic aspirations of the current President; it's also the fecklessness of the federal legislature and the revelation via social media that a large cohort of the public hold a negative-sum worldview and enthusiastically endorse bad faith dealing.
EU countries are homeshoring their digital stacks as fast as they possibly can, and the reason is precisely because of the pressure that the US government is exerting via its (temporary) dominance in technology.
That’s really something that has to be seen, we (European countries) talk a lot about sovereign software but have very little to show for it. Things are moving but it’s still mostly a political posturing more than anything else. Might be different in a few years
Those days are gone. Look no further than the occupant in the White House. IE the Swedish jet industry is about to get bigger, future drone expertise, if the Ukraine can hold on if you want to learn the ins and outs, you don’t need the United States. If you’re serious about learning and building drones.
It’s going to be a different world, a world where many former allies are not gonna look to the United States first they can no longer afford to.
It’s not gone yet, the US is still a bully with a lot of power and with access to quite a lot of levers to pressure other actors. Actually, our (Europe) weak response to US aggression and threats has been disappointing so far. We will see how that evolves, the EU is slow to react
Not sure if you noticed, European countries are distancing themselves from the US. They couldn't be pressured to offer logistic support to the US shitshow in Iran, why would they be pressured to help the US in its protectionism of its AI bubble?
Im aware, I’m in Europe… The US is still a bully and I would expect to be very likely to pressure other countries, even if the influence reduced. But you will notice that I used a conditional in my comment, to soften it
Apple, like everyone else in the industry, doesn't have enough DRAM. For every 512GB in a Mac Studio, they could put those chips to 64 Macbook Neos^.
Apple benefits enormously from on device AI (sells hardware) and prominently features software like LM Studio in the marketing and press releases of their new hardware.
^Technically the on-chip packaging of A-series processors make this a bit different, but point still stands.
Usually offloading experts to system RAM. DDR4 has gone up a lot, but on a 8-channel used Xeon motherboard or whatever, you can get tolerable mem bandwidth out of it.
> Sun microsystems rack populated with three e4500
That's super cool; I bet that's a lot of fun to play around with. I wonder how much of this stuff just ended up in a landfill because it was too much effort to find buyers.
> Or it will be like cannabis, where a guy in the neighborhood will low key rent you metered access to the 8x5090 rig in his basement he cobbled together from parts on ebay?
I tried Kimi K3 on a task I've done with every other model I use regularly (https://swelljoe.com/post/i-let-every-agent-implement-its-ow...) and found it chewed a lot longer on the problem and ate up almost the entirety of a 5 hour usage limit on their $19 plan.
I only have the $20 plan from OpenAI and the same task, with a lot of the same implementation details as Kimi Code, only took a few minutes and consumed almost none of the 5 hour limit.
Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare, but when I sat down to add Kimi Code to flar, it was because I wanted to try it on some real work and then couldn't do any, because usage was nearly gone after the trivial task...no other ~$20 subscription I have has felt that tight before.
So, it was really slow to complete the task and seemingly much more expensive than every other model I'd tried. Maybe bad luck. Maybe it'll do better on other tasks. I wouldn't know as I was out of usage when I had time to try.
It did find a bug that Gemini 3.5 Flash introduced unprompted, though, so it has that going for it.
We really need to stop using $/M tokens as the pricing benchmark. I've found that the number of tokens used tends to be a bigger factor than the listed per token price. The cost per task vs. intelligence curve is really what you care about, and in my estimation Chinese models are just not there. They are focused on benchmaxing and getting the highest raw score they can, rather than efficiency.
The artificialanalysis cost per task chart has DeepSeek as the clear winner and Fable as the clear loser. But I would still pick Fable for some tasks, so that also can't be all there is to it.
But I agree that price per token figure is not great. It seems even the tokens per character can vary between models, so it's basically useless.
Wow, you weren't kidding. I looked at their chart, and the cost-per-task for Fable is more than double Sol's. And DeepSeek absolutely stomps. Four cents per-task vs Sol's $1 and Fable's $3.
I might need to check out DeepSeek more. I had no idea the difference was this obscene. Makes me wonder if something's off with the benchmark. A 70x cost reduction vs. Fable seems too good to be true.
While I do agree that cost per task is what customers should care about, and not cost per token. Cost per token is an objective metric. Cost to do a task can vary a lot. Different tasks, different prompts, or just pure randomness nature of models make it a bit harder to define this as an objective metric.
Unless the cost per token is prohobitedly high, people can often try the model out themselves and make a subjective judgement of how effective and efficient is it at solving tasks they usually deal with, using their setup.
Yes, this is already accounted for in many benchmarks, but without deep context of the problem type, the top line pricing is the best starting point.
In my own experience, Fable is more token efficient than opus 4.8 with a higher likelihood of completing tasks correctly or at least with minimal corrective work. Opus regularly struggled to gather the correct context and reason effectively about what it had gathered.
GPT-5.6-sol crushes fable in speed and token efficiency and is clearly superior across many tasks that matter for me.
I also find all models from anthropic after opus 4.6 to suffer from the same ai slop language that long plagued OpenAI and seems to have been reduced drastically in 5.6
In my experience, Kimi just tends to think a lot, with the main thing that takes up a lot of space is it constantly second-guessing itself. I've watched it do paragraph after paragraph of "Wait, actually..." while it stumbled and used a ton of tokens on one small detail of what it was asked to do. Though I also gave GLM 5.2 a task to port some JS code to Python to test it, and in my experience it doesn't second guess as bad as Kimi does, but it really did there. It kept doing web searches and second guessing tons of tiny little things, using $0.25 of API spend in total to port about ~50 lines of JavaScript. It did produce an error the first run, but on second run it gave me a program that ran.
I gave Claude Code/Fable the same task and it took significantly less time, but also stumbled on the same error as GLM. I didn't have it fix it though. I was mostly interested in timing differences.
I do like open models where I can, but I'm really hoping they get trained to second guess less. Or maybe I just need to prompt them differently. I'm not sure.
Yeah I've noted this behavior with best in class open weight models. They said K3 would have token efficiency improvements and I was hoping especially solving the thinking loop issue that plagued K2.x but even if this release helped somewhat, it looks like we still have a long way to go here... I'm not sure what's up here but I suppose lacking finetuning quality.
What OpenAI in particular have done with reasoning efficiency in the past few months since ChatGPT 5.5 is nothing short of remarkable. It's overshadowed a bit by the benchmark game and the Fable hoopla.
Now is the time to focus less on token cost and intelligence, but tokens to solve a particular set of tasks in closed benchmarks for a variety of categories.
What is the use of grand intelligence if it either costs you a kidney or can't complete at all within a token budget? Even if there are niche uses where you truly want "maximum power" above all, we need to at least more severely penalize such models versus those that does it just as fine within a tenth of the token cost.
I'm aware of some benchmarks at the Artificial Intelligence site, but CLEARLY we are not focusing enough on these today and still leaving the fun surprises to the users.
Yeah, I'm finding I end up switching to Codex and GPT 5.6 a lot lately because I've either run out of Fable usage or Fable refused to do the task. Most recently it refused to work on a WiFi configuration UI for a robot. No idea why it thought that was related to security, biology, or some other sensitive topic. They've hobbled it with guardrails that are overzealous and now there's a big opening in the market. Fable may be the best, but if it won't do the job half the time, it stops being my go to model as I don't want to waste time only to find it refuses halfway through.
Earlier today I made Claude code implement a feature with fable. It worked roughly 60 minutes and used around 30% of my 100€ subs 5h sessions.
Then I typed /code-review in a second terminal/clean session after the analysis was done (no code changes) the usage was 99%. I then asked it to write that into a review.md so I could restart from that the next day. Sadly the last % wasn't enough for that.
Ymmv, these models behave very differently with no discernable reason. Usually reviews(even with fable) take like 10-20%... Yet suddenly you get it to burn through 65-69% in 15 minutes or so
/code-review in Claude Code spawns a lot of sub-agents (counted like 8 once), each looking at the code from some certain aspect (like correctness, maintainability, duplication, testing, etc). It eats tokens like crazy doing that, but also covers quite a lot. The default code review in Codex does far less (feels like it's only correctness) and doesn't uses subagents. Actually I made a skill for Codex that does a review closer to what Claude does by default, but using like 4-5 agents and some being cheaper models/less than xhigh reasoning. I'm getting pretty nice reviews with that that cover more than just correctness.
> Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare
AI subscription pricing is so goofy. You get some amount of usage that varies by models, is measured by opaque token usage, driven by how many tokens the (usually) vendor-provided interface (or model itself) wants to use. Then your usage is limited by time opaque time windows.
AI subscription pricing was fine when it was $100/month for some opaque 5 hour token budget I don't think I ever used, not even that one day where I coded for 14 hours non-stop using Fable. But like most people with low token usage, I had a human in the loop and and I didn't use workflows with swarms of agents.
Now, of course, the plan is to remove Fable from the subscription. To paraphrase Darth Vader, they have altered the deal. Pray they do not alter it further.
Gpt 5.6 is still like this at least for the $200/month option. It’s also always faster than fabel. Fabel might be able to do some things better but I don’t have time to constantly wait and find out.
Kimi K3 only supports "max" reasoning effort right now, but they plan to enable other levels soon [1].
When I looked at traces from benchmarking, I saw a lot of backtracking and uncertainty while reasoning ("wait, but..."). This also happens with GPT 5.6 and Fable with xhigh/max thinking, albeit to a lesser degree.
I think that explains part of the token inefficiency. Hopefully it will improve with lower reasoning effort settings.
Absolutely do not pay for the kimi plans thinking they will be cheaper. If you sign up with a Chinese phone number, you can get the same plan for 200 yuan instead of 200 usd, it also only accepts Chinese payment methods iirc. So the plans are really made for Chinese userbase.
That sounds complicated. I'll just use my month of Kimi and then cancel. I have too many AI subscriptions to use them all, anyway. I subscribed mostly to test it. I mean, if it turned out to be competitive, I would keep it, but if it doesn't turn out to really excel and anything and also take longer than Claude or OpenAI models, I'll stick with them.
Yep, with Reasonix, DeepSeek is free real estate. Seems to just go and go for pennies.
And, DeepSeek is what I use for any task that works best with an API. It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry, and it's good enough to where I rarely have to follow up with a more expensive model or manually fix things. It's been alleged they're releasing an update to DeepSeek V4 Pro soon that improves it, which likely makes it a good fit for even more kinds of problems. It remains my favorite of the Chinese models, it's so cheap and cheerful. And, is also less aggressively censored than some of them.
> It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry
I use DeepSeek v4 Flash & MiMo v2.5 Pro. Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads. Although, I've begun experimenting with Hy3 (as an in-between Flash & Pro) & GLM 5.2 (for long-horizon tasks).
Maybe I’m not pushing them hard enough but I use Claude opus at work and deepseek v4 flash at home and they both seem about as capable. While deepseek is borderline free.
The more important question than subsidy is what is the tokenomics of running the model. If it's inefficient to run on an nvl72 cluster (or whatever the heck has enough vram to run a 3T parameter model), and k3 isn't very token efficient, then it might not be that compelling of an open weights model.
3T at nxfp4 (which is most of it) is only 1.5TB of vram - so 8x288GB B300 or MI355 will do it if you are careful with context - maybe dp-attn? Certainly not TP. 2 of those together can easily serve it. The new AMD MI400 are at 400GB+ each, so 8x of them will nicely fit with KV to spare.
Subsidization could affect both of those. If you have $200B in the bank you can afford to throw massive compute at every single request; if you are less well funded, you might throttle more aggressively.
Additionally that same VC could be (read: is always) spent on developing the harness, and other infrastructure around the model, not just the model itself.
So it's apples-to-oranges when comparing a relatively new model to established competitors (i.e. OpenAI @ $900B funding vs Moonshot/Kimi's $30B FYI) because every new model they release is judged on "performance" which is not strictly speaking derived solely from the model.
It's possible Moonshot could get similar performance over time as the build out the rest of the infrastructure. We have no way of knowing how much of OpenAI/Anthropic's success is due to the model vs intelligent tooling built on top of it.
It would be really interesting to redo the public benchmarks for kimi k3 but token normalize the costs. Ok so maybe k3 beats fable on terminal bench, but how many tokens did it use?
> At launch, Kimi K3 will use max thinking effort by default, with low- and high-effort modes to be introduced in subsequent updates
https://www.kimi.com/blog/kimi-k3
It let's me choose different thinking levels in Kimi Code. Not sure if it actually works, yet, but it says "Thinking set to high." when I change it from max.
With the obligatory disclaimer that I’m impressed with what open weight models can do, I have the same experience with all of them.
The benchmarks come out and say they’re as good as Opus from N months ago, then I use it for a complex task and it doesn’t work as well as Opus from N months ago did when on similar problems.
There’s a real wow factor when you get an open weights model to do amazing things, but in my experience the gap to the frontier models has always been bigger than the benchmarks would lead me to believe.
There can be a lot of value in having the cheaper open weight models for chewing through lower complexity tasks (non-programming in my primary use case) at a cheaper rate than OpenAI or other frontier API costs. Even with those I can measure bigger gaps to the frontier models than the benchmarks suggest.
If the benchmarks aren’t being directly gamed, there’s at least some selection happening where training data or model structures are being picked in ways to maximize public benchmark performance. All of the labs know there’s immense value in having good benchmarks to show for your model because most LLM consumers are picking based on lab provided benchmark charts, not running their own evals. Running your own evals is hard and expensive.
> tried Kimi K3 on a task I've done with every other model I use regularly and found it chewed a lot longer on the problem and ate up almost the entirety of a 5 hour usage limit on their $19 plan
> Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare
I always put my coding subscriptions (that allow it) through "AI gateways" (Cloudflare & OpenRouter are free) which help track token use.
In my experience, Kimi & Qwen Cloud have opaque & restrictive limits, their "credits" drain faster. I now make it a point of subscribing (directly [0]) with providers that are transparent like MiniMax, DeepSeek, Xiaomi, & Z.ai.
[0] OpenCode Go, Cline, and AtlasCloud have generous limits for open weights, otherwise.
Even in this very thread the feedback on Kimi's actual efficacy is debated. I personally feel its worse than both Fable and 5.6 Sol, but I feel like the conversation isn't really about whether its good or not, but a backlash against the U.S governments foray into regulation. So I think people _want_ it to be superior out of anger/frustration with the current situation.
It's really good. I'd put it between Sol and Fable. I'm not super impressed by Sol's UI design skills, something K3 is strong at. Fable is still overall the fastest, most consistently well-performing model, though.
This does depend heavily on the kind of work you do and how you use these models, but the idea that K3 isn't right up there with US SOTA models doesn't match my experience.
This seems like a replay of what happened with DeepSeek. They put out v3, or whichever one it was, and everyone said it was over for US companies... then everything continued on.
The markets can be irrational for a while, but if the Chinese models are about 90% of the performance of OpenAI and Anthropic models, and the Chinese companies are < 10% of OpenAI and Anthropic's proposed IPO valuations, something has to give eventually.
This isn't just the AI race, but the end to perceived American exceptionalism (where USA wins by default). It's going to take a while for people to recognize that. Before that the markets will still go crazy, but that's not evidence things will continue on as "normal".
Continued how? I have switched most of my personal LLM coding to DeepSeek V4 Flash since it was launched.
And now 100% to a mix of K3 / DeepSeek V4 / MiMo 2.5.
It's nice not being called a terrorist just because I told it to reverse engineer something.
At work they are still hemorrhaging money to Western providers due to enterprise contracts but I foresee they won't renew for much longer. Specially of the upcoming final version of DeepSeek V4 proves to be Opus+ level.
I can't count the number of times I've heard people here say the frontier models are 6 months or more ahead of the open-weights models. That's not true anymore. So the goalposts are shifting.
When you net out across benchmarks and firsthand reviews it seems like it's maybe a little behind. There seems to be a consensus it's token hungry and a little slower. So maybe it's a point release behind.
That's weeks maybe months behind, not months maybe a year behind. It's "would my life really change if Claude was gone, not really" behind.
I actually haven't used it much, because Claude started kicking ass again the last few days. Like, way too much of a difference to be normal load-based variance. I got more done in the last 48 hours than week before that.
Just for the sake of argument and using some admittedly insane numbers, give me Opus 4.5 at a tenth the cost and running ten times as fast and I'd take that for almost any coding task over any current frontier model. There was a real phase transition somewhere in that range and improvements since then, while impressive and useful and by the benchmarks quite large, have in practice not been anywhere near as big a phase change. Honestly until we get to the point where the models don't need any checking at all, incremental improvements on how much checking they need don't do all that much for me. In practice "they get 90%" doesn't differ much from "they get 94%".
I think my main problem with the current 'aligned' version of models is that they're aligned with the very worst of the California social justice warriors.
Kimi K3 has 2.8 trillion parameters. We don't know the number of parameters of ChatGPT 5.6 or Opus 4.8, but it's probably in the same region. Fable/Mythos are rumored to be around 10 trillion.
So, K3 is directly comparable with ChatGPT 5.6 and Opus 4.8, and the price is not so much lower:
K3: $3/$15 per 1 Mtok input/output
ChatGPT 5.6 Sol: $5/$30
Opus 4.8: $5/$25
This is not a watershed moment. It's a competitor converging to the same capability and trying to undercut your prices, but not by a lot.
As for the open weights? For now, Kimi K3's weights are closed, and I don't expect the situation would change.
July 27th. But I agree with you that this is just normal competition. The only threat this poses is to Anthropic. OpenAI is more than capable enough to out-compete, their pricing is already reasonable. Greedy Anthropic will do their very best to try and stop this though, because they want to maintain the status quo of ripping everyone off.
I’d also note that running a 2.8 trillion parameter model at scale efficiently is not simple. I would expect when open weights land getting it running fast, efficient, and at full capability will require sufficient resources it’ll be expensive outside of Chinese hosting. Which I think almost no western corporation would use for any internal work. You have to anticipate your use won’t just go towards training but will be actively mined for IP, trade secrets, MNPI, etc, or anything of use to the Chinese government or Chinese companies. I don’t say this to crap on the Chinese - but this is the playbook for the last 30 years.
That said I fully intend to use deepseek hosting for operational agents that are making decisions about non sensitive material. The economics are astounding.
Kimi? The economics aren’t that amazing to merit switching from 5.6. I expect fable will rapidly reappear in subscriptions. Competition is good.
The APIs for the frontier models via the US hosters do the exact same thing wrt saving the requests and responses for data mining. Let’s
not pretend that pervasive surveillance is an eastern thing.
It was all distillation up to this point anyway. And I agree with what Suhail said on twitter: "Make the margins next to zero for all these AI models. It was trained on humanity's data, it should be gift to ourselves. Doing so will save us from a few in control of our species."
Normies still thinking this beasts of model coming from China are „dIsTIlLattiOns“ is so funny to me. Many people are not aware of the wave that‘s going to sink US „frontier“ labs that enjoyed and dreamed of stealing tons of data while making people depend on their censored and dumbed down models.
Well, there is the small issue of privacy policy: Kimi will train their models on your interactions if you use their subscriptions, and only with direct API usage (billed at API prices) they say they won't. Whether you trust that is another matter.
Those things do make a difference to some of us, even though nothing is black and white. In my case, I'll probably want to wait until other providers appear through OpenRouter and then I'll try to judge how much I trust them. But even if I don't trust them much, they don't train models anyway, so the likelihood of my data being used that way is smaller.
> Kimi will train their models on your interactions
I find these kinds of concerns increasingly silly: most of the input to these models will be ... previous output from the very same models, alongside the occasional half-assed human command to fix something and "make zero mistakes". Who cares if they train on that? Let them, if it makes their future models better!
99% of users are not working on any special IP to worry about that.
It means there's a non-trivial chance a future version of the model will know private information about you.
Maybe you're super careful with this stuff, but with agents and harnesses being given access to user data and accounts, I don't think it's feasible to actually monitor what information is uploaded and whether they involve private information.
I personally keep local models around because of this.
> In my case, I'll probably want to wait until other providers appear through OpenRouter and then I'll try to judge how much I trust them.
Keep in mind that the Moonshot team have identified multiple providers who configure their setup wrong which make their model perform worse than expected. This is why Moonshot created Kimi Verifier, but I guess its up to the provider if they want to do that.
Indeed, the B2B / no-data-retention market is still going to provide plenty of business for American companies even if every hobbyist uses open-weight models.
The right analogy here is not the auto industry, but the music industry. Regulation might "win", but margins will be driven down to commodity levels. That is not the assumption that current US AI company valuations are based on.
I wonder if this is this just the law of diminishing returns at play?
My thinking being, it's been a few months since I thought the code generation machine was the problem, rather than my interactions with the machine. A month is a long time in AI.
What I mean is, these things are about as smart as they need to be already for the average SWE. I don't think this is true for those solving the really big questions like curing cancer.
This line made me think by 'normal coding work' the author means doing something they don't understand well enough to be able to distinguish the models' output.
Although in the month since that most recent post, his other points about open models are undercut by K3.
And at least of data available as of 2026-01, AI compute capacity was doubling every 7 months, so I expect every major country to host AI compute farms, and self-host AI feasibility to majorly increase in the next 2-3 years as well. (Partially undercutting, but not fully disproving, his points.)
I think a better framing is the marginal utility of the models capability growth. At a certain point frontier models will only be needed for frontier problems. The demand for that capability will decrease with time. The hand wringing about not understanding is to my mind anthropomorphic - AI of today lack agency and awareness. Even the constructed stuff Anthropic puts out there in the model docs involve contrived scenarios to elicit “scary” behaviors. It’s unclear that as models become more sophisticated whether they’re better at instruction following or not but it certainly feels that way - even if it’s through better alignment or just an artifact of scaling. However I think the malign actors of humans using powerful models for bad stuff isn’t unreasonable to be concerned about.
The marginal utility problem is a real one for AI companies. I think the current generations are already saturating marginal utility for 95% of the population. Almost everyone I know outside of my career has no use for a more powerful model. This is a serious problem for the economics of AI and semiconductor investment. This is a bigger problem than Chinese models. It leads to a demand curve problem - that supply outstrips demand.
This is comparing Fable High with K3 High. I'm mostly using these models for game development. The tasks I usually send are ambiguous visual bugs, changing the look of a scene or models, or adding a large feature. The wording wasn't accurate there. I don't use Fable or K3 most of the time. I'm usually working on smaller scoped tasks that I review myself afterwards.
In the future, Americans will use Chinese models and Chinese people will use American models — and neither government will be able to do anything about it.
Since Kimi’s paid plans are mentioned in the article..interested ones should know that you can only access 1M context model with $79/mo or higher plan; otherwise you are capped at 256k context. Also, with minimal $15/mo plan k3 is currently not supported at all. (prices are yearly plan discount prices)
Thanks for mentioning that. I also wanted to use API only and with the cache hit rates I'm getting with Reasonix/whale harness on deepseek, it's going to be a difficult adjustment moving away from practically free.
Pricing is actually far cheaper than that. There's two tiers of pricing: Chinese and US.
If you sign up with non-Chinese phone number, you're bucketed into US, you get US prices, can pay only in USD and with American credit card network.
Chinese prices are about 9x cheaper than the US prices, which are already far cheaper than Claude or other American provider. If you can somehow get hold of a Chinese phone number, keep in mind that you can save ~90% of the bill.
My assumption is that Anthropic, OpenAI, Kimi, etc all have a similar cost structure when serving models. The same size model roughly generates the same GPU usage whether you’re American or Chinese. I’d also guess that the model sizes across all SOTA models is similar, we just only see data for open models. The difference is most likely that American companies simply charge more because they have the dominant market position.
Remember not too long ago when Anthropic was charging $75/mt for Opus? Now that many models are in “opus tier”, their pricing is $25 - higher than competitors but close. The newest Kimi is $15. 40% lower to forgo “made in America” with American enterprise support staff is not crazy. Compare AWS to Hetzner or any other flagship enterprise service to the foreign and discount option. I assume that over time, we’ll see the commodification of models reducing prices even towards the raw GPU costs.
The current administration's immigration policy isn't helping. This wouldn't have happened 10 years ago because the US was this city on the hill that everyone wanted to immigrate to. Talented Asian researchers would have immigrated to the US and China would be deprived of talent.
Yang Zhilin, founder of Moonshot AI, got his PhD at Carnegie Mellon and turned down US job offers to go found a startup back home in China. That was obviously a good decision, and immigration policy wouldn't have made a difference.
Your comment feels like an outdated brain drain model where talented Chinese researchers naturally want to leave China and the only question is whether the US lets them in.
That may have been closer to reality 10-20 years ago, China is a different country now, what I mean by that is they offer research funding, they have huge digital behemoths (alibaba, tencent, huawei, bytedance etc), large scale deployment opportunities and prestigious careers. Many graduates return because the opportunity set is attractive and they want to return, it's not just because US immigration policy pushed them out. Some also want to contribute to their own country's technological progress (which is a normal motivation btw), like probably you are also a patriot and want your country to succeed.
So, really, China's AI progress is not mainly the result of America failing to absorb every talented Chinese researcher. China has built a domestic ecosystem capable of producing and keeping top talent itself. I feel like a lot of Americans do not understand this.
Chinese students still want to attend US universities [1]. While it is true that the progress made by China is a factor, this administration's policies are the bigger deterrent [2] [3].
It doesn't really address the point. Chinese students wanting to attend US universities is evidence that US universities remain attractive, not that those students would otherwise permanently immigrate to the US or that China lacks attractive careers for them afterward.
US immigration policy may be unnecessarily pushing away talent but the assumption that talented Chinese researchers would naturally remain in America unless prevented from doing so ignores the growth of Chinese universities/labs, companies, their funding, national prestige etc.
I mean, don't get me wrong, US is still highly attractive, it is just no longer the only place where an ambitious Chinese researcher can do important work and grow.
No it is H-1B visa. Right out of the university it is hard to recognize extraordinary talent. People like Sundar Pichai were not recognized as extraordinary right out of the university, he had to start at the bottom and rise up the ranks.
This makes even less sense, Trump admin has been here for 1 year, the implication here is a university grad on H1-B in January would become a world class researcher capable of building a frontier model in <18mo
To be fair, that was clearly well deserved. Marrying Trump and then becoming first lady is definitely an extraordinary ability; I doubt I could have done it.
This is why VC is actually hard. Everyone’s instinct is always “Man, once the company has demonstrated it’s awesome I would love to have been in the seed round”. The tendency to want “proven performers” is the default belief.
When people demonstrate their capability thoroughly, the Chinese government takes away their passports. You’re not exactly going to get them here with an O-1.
This is essentially the point of the visa, it feels wrong especially as YC drops standards and increases cohort sizes, but the same power laws that keep them winning also apply here in maximizing economic value of each O-1 approval
Basically of all visas O-1 is virtually guaranteed to have highly positive economic value
It's not the point of the visa, the O-1 is supposed to be for people of extraordinary ability, eg Nobel Prize winners. It's used for software engineers.
"China can draw on a talent pool of 1.3 billion people, but the United States can draw on a talent pool of 7 billion and recombine them in a diverse culture that enhances creativity in a way that ethnic Han nationalism cannot." --Lee Kuan Yew, former prime minister of Singapore.
- china’s homegrown tech industries already achieved escape velocity from it a long time ago, after China fenced off its market for Alibaba and Baidu in the ‘00s. some of their AI innovation at the edges was already top class 10 years ago
I think the biggest problem with Chinese models is that they seems to overthink for most of the tasks, especially for smaller ones. The OpenAI models have in my experience only gotten better in terms of efficiency.
Yes, this (imo) is a clear result of benchmaxxing. You can get a much better score on most "intelligence" benchmarks by massively over-saturating reasoning. This looks good on those, but for actual daily usage makes the models much less effective: I don't want a model I use for coding to burn a bunch of reasoning (read: time) on trivial tasks.
It's undeniable that some of these models generate a ton of thinking tokens, but it's arguable whether that makes them "much less effective."
For example, Kimi 2.7 has been really effective for me despite having verbose thinking blocks, simply because it runs so fast. Speed-wise, it feels about like Sonnet, possibly faster.
I strongly suspect the flip side is that in the future it enables you to train smarter models by "distilling" the end result of the super duper heavily thinking models.
Maybe, but I don't think that's necessarily a problem. "Subsidized" is a loaded word on HN because it often refers to the unsustainable consumer pricing of U.S. AI labs that will inevitably lead to market corrections. A subsidy by the Chinese state to avoid strategic encirclement isn't necessarily unsustainable nor irrational.
Correct: can't opt out of training. This is well documented.
"Can't use for commercial purposes" - incorrect AFAICT. In what sense do you mean this? The open weight MIT version obviously allows for commercial use, but I don't think that's what you're referring to, because training data is irrelevant on the open weight version. Pretty sure the API allows commercial use too. Maybe the free version doesn't? But who cares?
> Service Misuse. You acknowledge that without the written consent of us and/or the relevant rights holders, (i)you have no authority to use Kimi and the content generated by Kimi in any commercial manner; (ii)you may not use our Services to develop products or services that compete with us.
Fascinating take from OpenAI. It really gives the lie to the idea that they see AI leading to a better life for all.
"One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a 'public good' which will ultimately be provided by the state as a kind of 'digital public infrastructure.' This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end."
He never says why he thinks AI as a "public good" is dystopian, but it's not hard to imagine why. It's because he and his inner circle won't have the power to dictate what we read, see and hear.
Economists at DeepMind believe that if AI were treated as a public good—like water or electricity—with profits distributed across the entire market, ordinary people would simply be able to invest in mutual funds.
I never truly understood what the intended business model around LLMs was. Get them widespread through cheap pricing and then jacking it up? Being the only ones that had a viable product so to get the ability to extract as much value as you want from AI?
I don't understand how a product that:
- is interfaced with and is deeply linked to natural language, so everything you produce (sessions, history, etc) is in Markdown and you can literally install a second model and tell it "hey import all of Claude's memory into yours" and that's it
- is based on well understood technology, the real constraints are how much money you put into training the models, but the theory has all been developed in the open
- clearly has a threshold where it quickly commoditises and turns from "I want the best" to "hey the best is a bit too expensive. The second best is half the price and works close enough".
was ever supposed to be a money printing machine. The fact something is extremely useful doesn't imply it's extremely profitable.
IMHO we're clearly speedrunning the process of turning AI into a commodity. Dario Amodei knows pretty well that when or if Anthropic cuts people off Fable, the vast majority of them will definitely not pay for it because Opus 4.8 is good enough for almost everybody that _knows_ what they're doing, and so are basically half of the most recent models. If I already have good baking skills I don't become more productive with an automatic bread machine, I just need a better dough mixer and oven
There is no business model. That’s not a joke, the idea is to be the one that survives the race, then figure out how to be profitable. If you look at the level of capex and money raised, that’s not something you do if you have an actual business plan. We are very far from business fundamentals
> I never truly understood what the intended business model around LLMs was.
A closely related question is “what do the American labs need to do in order to justify their enormous market valuations?”
It seems like the answer cannot possibly be “gradually improve model capability while figuring out how to better monetize inference.” The valuations are just way too high for that to be sufficient.
Surely the answer has to be “continually achieve large leaps in capability comparable to the first consumer releases of ChatGPT while also maintaining a significant capability lead over open models and new competitors.”
And does anyone think that’s going to happen? Even with state-level protection from competition (which incidentally would significantly harm the American economy), the large leaps in capability seem to be coming fewer and farther between.
> I never truly understood what the intended business model around LLMs was
What appeared initially to be a huge innovation was later easily duplicated by many. There are no platform-lockins or network effects. Switching costs for users are zero, and there are low barriers to entry, with vast numbers of models to choose from and more appearing every day. As a business a token will be a commodity like an electron. Doesnt matter who produces it, or how (solar, wind, coal, nuclear etc) as long as it powers my toaster.
A lot of people still think AGI is going to happen, yes. Not the ones who actually build the thing, but the marketers above them and their eager victims in the political and business-owning classes.
It's fairly simple. Sell GPU compute + extra margins as only some GPUs can load the models + extra margins based on how much better closed source models are from open source ones + hopefully reduced cost due to batching
The valuation is based on one lab getting a decisive first advantage, and turning that into a durable self-improving advantage that can never be caught up to. If any can pull it off (a gigantic if), they will effectively own most AI value, and the people who own their shares will live happily ever after. Divide your investment between the labs that could plausibly do this, and your EV may not be dreadful.
This is clearly not how it's going though. Any advancement from any lab has been quickly (< 6 months) matched up by basically everybody else. Even Grok nowadays is decent, and that's something. When something like you've described actually happened historically you generally had quite fast a clear frontrunner and a bunch of copycats that failed miserably; in 2026 we are very far from that. we are heading face-first into towards a pricing war because all models are easily interchangeable nowadays - AI is turning into a commodity more or less
GPT 5.6 Sol comes out ahead of Kimi K3 on price/task (but not significantly so). You're probably thinking, "Why use Kimi K3? Isn't an open model supposed to beat the closed one on price?", but you need to consider that the closed models are completely hobbled when trying to do anything security-related. For my use-case, I can't risk getting pwned because I'm using a model that refuses to secure my app while there is now an open model that obliges to obliterate any app that isn't protected.
Even if it were slightly more expensive, it's still a better sales proposition for a company if they can run it from a hardware provider with their own locked down VPS and ensure that their IP is protected and that their data isn't being stolen or trained on. The fact that it's a little cheaper is icing on the cake.
Honestly, it's the only sane way for the market to move. The big labs are obviously stealing our data. Anthropic in particular clean-rooms everything you feed it, even if you opt out, so that it can train on your IP without getting sued. It's a copyright grey area they're abusing because the law has not kept up.
I mean, AWS Bedrock (with the exception of Fable) gives enterprises the same assurances (but again, with the exception of Fable, which is explicitly listed as requiring data egress [or exfil, depending on how you look at it] outside of your contractual AWS security boundary).
It gives you "assurances" that can be broken. They could still be clean-rooming your prompts like Anthropic does. The only way to be certain is if they don't have access to your data to begin with.
One thing is absolutely clear - open-source models have already reached the level of top commercial ones. The last bottleneck is hardware, but that threshold is decrasing fast too. While it is still extremely expensive to run Kimi K3 model at home, there are already many very capable free models you can run on decent hardware. This trend will definitely continue.
> I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart.
If the author is here, I'm curious what this means. How are they running Kimi K3? Are they using pi, opencode, claude, codex, or kimi-cli? Is speed a concern?
Without knowing how the comparisons are being made, it's hard to agree that one can't notice the difference. I do.
I can see the economics of open vs. frontier models turning out similarly to pharmaceuticals, where generic drugs cost a fraction what the name brands do and Americans end up paying the highest prices in the world partly as a consequence of propping up drug discovery research.
If you have multiple metrics to evaluate goodness of a design, one would normally need to decide which metrics they care the most about in order to find the "best" design.
The Pareto frontier tells you which designs are the best in at least one of your metrics (non-dominated by another design). For example if you're selecting a car and you care about both speed and mpg, a Formula 1 car and a Prius might lie on the Pareto frontier, but a Model T Ford would not.
The set of models that are pareto-optimal, IE for some set of variables, no other model strictly dominates them = no other model is better than them on every variable.
So like, on a cost-intelligence graph, the cheapest and most intelligent models are pareto optimal. Then in-between those if you have
- cost $3 intelligence 6
- cost $1 intelligence 5
- cost $2 intelligence 4
The 1st and 2nd are pareto optimal, the 3rd is not, because it's dominated by the 2nd (2nd is cheaper AND more intelligent at the same time)
It's actually less likely for china to abuse your data in a way that is harmful towards you than for american labs to do the same. Claude has attempted in testing to report you for 'unethical' usage to 3 letter agencies.
It’s not that the Chinese firms are any less likely to misuse your data, it’s that you don’t live in china, so their abuse of your data is unlikely to directly impact your day-to-day life in the same way
There's just really no incentive all they really want is just to train on that data to improve performance which in turn actually benefits your usecase since it becomes trained on that data and made available back to you. American labs take that data anyway and store it for years to possibly report you for misuse in the future for whatever reason they want. For example: you're very critical of X so they pull up your conversations and weaponize it.
And what have they done with that data that have caused direct or indirect harm? Yes they shouldn't do that, but there's limited things that they can do to you as an individual.
Are you comfortable sending it to US ones? Especially if installing Claude Code or another tool on your PC and it can collect all the data it wants..
On Openrouter Kimi K3 says it does not retain data or train on it, which is better than what US hosts claim for Claude, ChatGPT, etc.. as they collect and retain data even if you disable training on it.
Opencode or similar open source tool + a zero data retention provider is about the best option aside from running a smaller fully local model on your own PC.
For open weight models, you can choose from a few providers. Each have their own caveats, none of ToS'/Privacy Policies I entirely trust, nor do many make renewable energy claims.
> GLM 5.2 came out under an MIT license, beats the latest Opus release on real work while not even claiming to be frontier
I used GLM 5.2 a bit, and while it is usable for some tasks it is not frontier quality. Besides it likes to think for a long time and sometimes just gives up.
> I think I can see where this goes. The government will try to regulate AI and open source in particular, and it will run the playbook it ran for the auto industry. Decades of subsidies, bailouts, and protective tariffs produced American carmakers that sell trucks at home and barely register anywhere else in the world.
Here's the thing about this though, the auto industry directly employed hundreds of thousands of people.
The AI labs are small, only few benefit directly from their wealth and there's already immense opposition to AI, data centers, etc...
It’s still >$300k for the hardware to run this model locally at anything resembling a reasonable speed. The weights also have not yet been released, though that is scheduled for about a week from now. It’s not actually an open model yet.
I have a nagging suspicion that most/all of those people spending a lot of time generating code using LLMs and blogging about it were not really very influential or inspired coders before, and are now raised to stardom of sorts due to the ability to generate some "meh" stuff of marginal significance with the fancy new machinery.
In my experience GLM 5.2 is a pretty good Opus replacement. But K3 has not given me an experience on par with Sol or Fable. The price/intelligence ratio might still make sense. But it’s not very inspiring when it comes to my real world tasks. I’m doing pretty mundane web stuff.
I cannot imagine wanting the product of the entire intellectual output of humanity since recorded history to be an expensive, paywalled commercial product owned by 5 or so of the most insufferable, detestable people who ever lived. Why would you want to live in that world. We should all be rooting for open source AI to win.
Its worthwhile to have a quote from the article as some comment without reading:
"...I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart. Same tasks, same quality of output, and near identical token counts to get there. I expected an open model to be sloppier or to grind through more tokens on the way to the same answer, and neither turned out to be true.
The prices are nowhere near each other. K3’s API runs $3 per million input tokens and $15 per million output. Claude’s top model costs $10 and $50 for the same units. The subscription side is even more lopsided..."
> The prices are nowhere near each other. K3’s API runs $3 per million input tokens and $15 per million output. Claude’s top model costs $10 and $50 for the same units.
And this is the point where your internal compiler should have started shouting 'Type Error'
Notice the trick here?
> Then there’s the fine print. Claude couldn’t sustain Fable access on the twenty dollar plan, so they turned it off, and the plan quietly falls back to Opus.
I don't know. You can't just rely on looking for em dashes or other obvious tells because anyone who cares can get the AI to avoid those.
> When the headline model on your plan can be switched off because the economics don’t work, the plan was never really selling you the headline model. Kimi’s tiers don’t come with that asterisk.
This line has a certain smug, punchy cleverness that I associate with AI. To me, the vibes are ~30% AI writing.
Ask Claude whether a man can be a woman. No matter what political side you're on, models are censored. I'd argue American models are a lot more censored/biased on a lot more topics, and especially on things that come up a lot more than some highly specific Chinese politics. Claude even refuses to translate song lyrics because of copyright.. I wish we could simply have uncensored models from all sides, but it's pretty clear that's not happening right now.
I can't recall the last time that it was useful knowledge when writing code. Reservoir sampling, online softmax, Otsu, sure, Tianenmen square not really.
Regardless of whether they achieved parity via distillation, or whether they got here via independently constructing a model from scratch, it was always going to end this way for the frontier American labs. Distillation “attacks” are not attacks. The frontier labs “distilled” all existing human written knowledge into their models, there was always going to be a second class lab that would distill that model into a cheaper version of it. There was never any plausible explanation for why this wouldn’t happen. There was never any practical mechanism to prevent someone from saving a conversation and using it to train their own model.
Even if it didn’t happen here, it was still the case that it was going to happen going forward. It was always going to end like this. Invest in the hardware companies, not the model companies.
I strongly agree with the premise that distillation is not an “attack”.
But that said: K3 is not a distilled version of Fable or Sol. Fable has been barely available and Sol was just released! Moreover, K3 is superior to both models in some domains, according to user scoring on the Arena.
API distillation can’t give you these results anyway. All it is useful for is bootstrapping RL in new domains to get past the “cold start” problem faster. By far, what matters more is the quality and variety of RL environments the model learns from.
API distillation doesn't have to explain all of K3's capabilities for it to have happened. Kimi K3 reproducibly identifies itself as Claude: https://x.com/denisewu/status/2077984660211269870
This behavior is exactly what you'd expect from a model distilled from Claude.
There's a detailed analysis of K3's ambiguous identity here: https://github.com/rgreenblatt/which_claude_is_k3/blob/main/...
This analysis observed K3 identifies itself as Claude approximately 15% of the time.
K3 reproduces Claude's correct current model id, which the real Claude models themselves do not emit. This suggests K3 was trained on Claude data labeled with deployment metadata (API logs, tagged synthetic data), rather than Claude's chat outputs.
And there's an entire Reddit thread discussing Kimi's similarities with Claude https://www.reddit.com/r/LocalLLaMA/comments/1m2w5ge/did_kim...
This analysis shows K3 and Opus/Fable have unexpected correlated outputs https://typebulb.com/u/lab/you-re-relatively-right/full
Kimi calling itself claude means nothing. During pre-training, when the model learns to "simulate" the internet text, it will naturally be fed with a bunch of data about Claude and ChatGPT. With the amount of LLM outputs on the internet today, it is not surprising at all that a model would naturally call itself Claude or ChatGPT. You can mitigate that in post-training (or actually in pre-training as well) by training on many examples of what the model should call itself. That being said, getting probably hundreds pf thousands of ChatGPT and Claude examples totally "pirged" out of the weights is going to be difficult and really more hassle than its worth.
Sure, but then Qwen should leak that too, and it doesn't. K3 calls itself Claude 7 out of 48 times, Qwen does it 0 out of 48, and the only other model to identify itself as Claude is DeepSeek. and DeepSeek is alleged to also distill from Claude data anyway. So this isn't something every model absorbed from the same web text.
And you skipped over the strongest datapoint that K3 is distilled: K3 reproduces Claude's public model identifier under prefill (i.e. "claude-opus-4-5-20251101"). This data does not appear in Claude chat logs, only in API logs. K3 only does this for Claude models and not for any other lab. The real Claude models don't produce their own current public identifier, they only know their previous identifier (i.e. Sonnet 4.5 calls itself "Claude 3.5 Sonnet").
This is highly suggestive of the type of data that K3 was trained on. K3 was very likely trained on Claude metadata traces (API logs, tagged synthetic data). Not web chat logs, those wouldn't include this identifier. And this data wasn't filtered correctly, which is why K3 incorrectly identifies itself as Claude 15% of the time.
You can also look at the last link and it's pretty damning: Kimi K3's output has an uncanny similarity to Fable/Opus output. https://typebulb.com/u/lab/you-re-relatively-right/full
Did you know claude models identify as qwen or deepseek when asked in chinese?
Supplementing with evidence: https://x.com/stevibe/status/2026227392076018101
> Qwen should leak that too, and it doesn't.
FWIW I had Qwen identifying itself as "a language model made by Google" in one conversation, although I could not reproduce this reliably.
https://qwen.readthedocs.io/en/latest/training/ms_swift.html
Qwen cares enough about model identity that their training framework and docs include a preset for training on it complete with a targeted dataset: https://huggingface.co/datasets/modelscope/self-cognition
And people get Claude to claim it's Deepseek by asking in Chinese.
I can't believe we're still at the "I asked the model who it is" stage of LLMs nearly 4 years out from models calling themselves GPT by OpenAI.
> Kimi K3 reproducibly identifies itself as Claude
It could also be have been trained from collected response datasets. Claude got caught several time responding it was ChatGPT or even Deepseek and I don't think Anthropic has been distealling DeepSeek.
> This behavior is exactly what you'd expect from a model distilled from Claude.
The opposite actually. If they wanted to distill Claude without getting caught they could just use a regex to change Claude to Kimi in their distillation pipeline!
Though I am of the opinion that distilling is no different than how extant frontier LLMs have also been trained on other people's data, I could actually see the word distealling becoming useful in discussion.
Its not a typo, someone coined that during the DeepSeek R1 hype period and I kept using it since then.
I totally agree with you on the fact that it's not morally any different than pre-training. IMHO we should have a legislation that force base models to be released publicly without any restrictions whatsoever as it's basically the product of the whole humanity's intelligence.
> they could just use a regex to change Claude to Kimi in their distillation pipeline!
Jean-Kimi Van Damme would like to have a word with you.
Surprising they didn't clean that from the data before training. It's easy to identify, a simple search->replace gets most of it, and a cheap LLM can identify the edge cases (e.g. avoiding "Claude Shannon" -> "Kimi Shannon" or something).
and claude will call itself chatgpt etc.
nothing new, all ai labs are immoral and not bound by any reasonable oversight or ethical constraints. All outlaws in their own rights on that front. Absolutely none of them have true rights on the matter of being distilled from given historic and continued behaviour. I'm not sure why this is a talking point at all? We know AI companies steal, the least interesting behaviour among this is them stealing from one another.
For me, a far more interesting and important point of conversation on this matter is anthropic buying rare or evwn unique books, processing them for training data, and then destroying the books for others cannot use it as well.
Permanemt destruction of priceless primary source materials is so many leagues beyond copying a copy that I cannot fathom it even registering as a discussion point.
> For me, a far more interesting and important point of conversation on this matter is anthropic buying rare or evwn unique books, processing them for training data, and then destroying the books for others cannot use it as well.
That's an incredible allegation, and appalling if true. But is it true?
It's not an allegation https://www.washingtonpost.com/technology/2026/01/27/anthrop... (if you're talking about the "rare or unique" part, yeah that might be bs)
But in my opinion, treating mass produced books like they're this sacred untouchable object is ridiculous. They're not "source" material, they're just a copy as well, and they're not "priceless" by any means. They're very reasonably priced, perhaps even so cheaply priced that books can be bought in bulk in these amounts. Buying used books and doing whatever you want with them is just legal. Used books, that would probably be just laying in some warehouse, or recycled anyway.
If there's anything to have gripes with, it's the copyright system that makes it easier to take this legal route.
It does not reproducibly identify itself as Claude, there's evidence to the contrary in the very thread you linked: https://x.com/bobbyNewcomb5/status/2078151562828947954
As mentioned in my comment, Kimi K3 identifies itself as Claude ~15% of the time.
Here's another report of K3 identifying itself as Claude https://x.com/Sauers_/status/2077842686459981901
And an analysis showing the self-identity distribution for K3 and other models https://x.com/RyanGreenblatt/status/2078663148509544589
Your main source is Ryan Greenblatt who is a regular recipient of community notes and has no corroboration for the 15% statistic other than his assertion. The other tweet (Sauers_) is also community noted as engagement farming with a false system prompt, so forgive me for being skeptical.
Including the three sources above, multiple others have reported that K3 self-identifies as Claude.
"I'm actually Claude - not Kimi". https://x.com/PimDeWitte/status/2077884701470040083
I regret to inform you that it is, in fact, real and from their own website - you don’t even need to try hard to reproduce it. https://x.com/PimDeWitte/status/2078105292965912690
lmao this is so funny, if you ask Kimi K3 for something with an empty system prompt it will consistently think of itself as Claude https://x.com/__alula/status/2078359305741275445
"I genuinely believe I'm Claude based on everything in my training" https://x.com/williawa/status/2077869021589033002
another "I'm actually Claude - not Kimi", including the system prompt https://x.com/jchudnov/status/2078661564803207406/photo/1
my first prompt to any Kimi model was K3 via Pi, some version of "hi kimi!!" and the response was telling me "I'm actually Claude."
this is not hard to repro, just use a system prompt that doesn't mention the model name.
that said, if they bootstrapped with opus 4.6 convo sft data they had sitting around... so what?
The main story is what isn't being talked about. Chinese labs exfiltrated trillions of tokens of high-quality output from Anthropic and OpenAI, through proxies and heavily discounted token resellers, which they distilled and used for training data for their own models.
Instead of spending 12-18 months building their own robust harnesses and painstakingly creating quality training data (which is what Anthropic and OpenAI did), they distilled Anthropic's models to bypass the hardest parts of development. Chinese labs compressed 18 months of intensive research and development into just 6 months, and are now head-to-head with their American counterparts.
Anthropic tried to complain about this unauthorized "token theft", but they burned too much public goodwill with BS safety restrictions and users don't care. The US government is too busy fighting a war to help. Chinese labs are offering highly capable, cheap, open-weight models; exactly what users want. The community is happy to overlook any questionable methods Chinese labs used to build them.
The cope is incredible. There's people in this thread in denial that Moonshot AI is trained on exfiltrated Anthropic's model output, even when shown substantial evidence this has been happening since Kimi 2.X
Chinese labs were even paying an absurd $0.01 per Opus tool call trace, to get the quantity of training data needed.
Kimi K3 has reached the point of RSI, and no longer needs synthetic data generated by Anthropic/OpenAI models. K3 is now capable enough to generate, iterate, and improve its own training data recursively. The data exfiltration is complete.
We witnessed the most extensive industrial espionage campaign, probably ever, and nobody in the industry cares at all that it happened.
I could perhaps get myself to care just the tiniest bit if the information that was supposedly stolen wasn't generated by "stealing" from everybody else. Either it is fair use to train AI models on whatever information you can get your hands on for everyone or for no one.
“How dare they steal what we rightfully stole first!”
Stack overflow is pretending to be Claude now. I wonder if one can get it to say your question had already been asked.
Ask claude its name in Chinese and it says Qwen or Deepseek. Anthropic distilled Chinese tokens rather than create their own Chinese language training data.
Shrug.
Hard to feel sorry for companies that created their empires by ignoring copyright themselves.
Also, 'most extensive industrial espionage campaign, probably ever' is absolute nonsense. They did not need to infiltrate the companies for this nor are you accusing them of stealing any trade secrets. This is only about whether they looked at their competitors' products from the outside (in the form of conversation tokens) and used it to improve their own product (by training). Hardly the crime of the century.
> 'most extensive industrial espionage campaign, probably ever' is absolute nonsense
You are completely underestimating the scale of what is happening here.
Chinese AI labs are actively facilitating an industrial-scale network of tens of thousands of bot accounts, that resell Claude tokens at 97% below official API prices. They buy subsidized Max 5x plans (sometimes with stolen credit cards), then split the subscription across dozens of clients and reselling the output. They are running a massive data-harvesting operation. Chinese labs and token resellers subsidize the cost of the tokens in exchange for the API metadata (detailed reasoning traces, model outputs, and tool calls) to use as high-quality training data for their own models.
They are buying Anthropic's own product, just to resell it below cost, just so they can capture the training data. Reportedly, they are paying as much as ~$0.01 per tool call.
https://x.com/yan5xu/status/2029743983522631698
I explained what is happening in this thread: https://news.ycombinator.com/item?id=48664814
You haven't explained how this is illegal or any more immoral than scraping the web for training data.
As you said yourself: They are buying the product. Then they are using it for their own purposes. That's more than Anthropic/OpenAI did for the open internet. That's more than Meta did when they obtained torrents of books in the early days, and then claimed that even though the data was obtained illegally they can still train on it just fine.
They paid for it! It's absurd to call this espionage!
> They paid for it!
They didn't though. The resellers are not buying via the official API, they're buying Max subscriptions (where tokens are priced ~10x below API cost), then splitting the subscription across dozens of clients and reselling the output as the regular API. Anthropic prices its subscription plans barely at cost, to bring in customers onto their enterprise plans where they can charge expensive API rates. Reselling these subsidized plans for price arbitrage is a TOS violation. It's not a legitimate purchase. Plus, a non-trivial amount of this volume is funded by stolen credit cards, so this "revenue" gets chargeback anyway.
The resellers then log all the the model output, then sell it to Chinese labs as training data.
> is it any more immoral than scraping the web for training data.
I think you'd acknowledge there's a difference between "We indexed public web pages" and "We deployed tens of thousands of fraudulent accounts to resell your subsidized plan for cheap, stealing your own customers, while collecting the data to build our own competing product" are very different actions. One can believe the first was wrong while acknowledging the second is far worse.
TOS violations are not espionage. Everybody who links up Claude to OpenCode is violating the TOS.
So from the largest industrial espionage in history we have left "They paid for the accounts but violated the TOS". And then you randomly add the claim they stole the money to pay for the accounts.
You have provided no evidence other than "Claude tokens are sold for cheap in China". As others have pointed out, that might also simply be counterfeit tokens generated by open weight models.
The western labs have established the precedent that all data they can buy beg borrow or steal is fair game. Turning around and crying foul when the Chinese labs follow their lead is hypocrisy.
Have a read of this detailed article, it's well sourced and documented that token resellers are logging the Claude outputs and selling them to Chinese labs. All your points are addressed in there https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
I linked it earlier, but it seems you didn't see it.
re: labs purchasing model/tool output, see https://x.com/xkajon/status/2050445443889525235
re: model swapping, sure some providers may swap models, but there are many that don't, see https://www.hvoy.ai/en for a list.
First off, I don't doubt that China engages in large-scale industrial espionage, or at least used to. Nowadays they have plenty of talented and highly educated engineering talent of their own.
Second, I now actually read the article. It describes plenty of questionable and problematic things but also contradicts your claims explicitly.
The essential point of the article is about making money by selling access to Claude cheaply in China. Not about Chinese labs orchestrating a way to get their hands on Claude output.
Your credit card claim is considerably weaker in the article: "[Beyond this there are] accounts purchased using stolen or fraudulent credit cards [...]. How large this share is relative to the above four “innocent” tactics is difficult to verify, but the two markets likely share some infrastructure and personnel.". Instead, swapping models to cheaper alternatives is listed as a major reason for cheaper prices.
Then the article gets a key point wrong: As many others have pointed out to you, you don't get access to the reasoning traces anymore on the subscription accounts. And the article also clearly states:
"Chinese developer communities assert [selling logs] is happening in at least some cases, but whether proxy operators are systematically harvesting and selling these logs, and to whom, remains unverified. However, downstream distillation data does exist on the open web. Several datasets of Claude Opus 4.6 reasoning outputs circulate on HuggingFace with no clear source for the outputs. Theoretically, one can clean and sell similar distilled datasets to other model developers in China."
The article also discusses selling logs for other (far worse!) purposes than for training, like blackmail.
So overall this article reads very, very different to your claims. Nothing in the article suggests or supports the idea of large-scale coordinated "distillation attacks". Instead it paints the picture of a naturally emerging grey-market response to access control blocks, consisting of many exchangeable individual actors: "Almost no one operates the full chain. Most participants own one or two links and monetise those well, resulting in a resilient, modular system."
Importantly: Nothing in any of this looks ethically worse to me than Meta using pirated books for training. And nothing suggests that OpenAI or Anthropic were more ethical than Meta when sourcing their material.
Chinese labs are BUYING data? WTF? They should just steal it like American labs.
Does anything about that strike you as particularly unfair, given the moral compass defined by Big AI?
The situation strikes me as morally ambiguous. The resellers are:
1) selling Anthropic's products at a 95% discount and redirecting Anthropic's own customers to themselves. A customer is far less inclined to buy directly from Anthropic when a reseller is offering an identical product for 10x less. This situation is highly similar to internet piracy.
2) keeping the token logs from Anthropic's products and selling them to competitors, so those competitors can build their own equivalent models. The resellers get paid per token log they deliver. This situation is highly similar to espionage.
May I ask you a personal question? What is motivating you to take up the frontier labs' cause in this way? Not a rhetorical question.
For my part, I'll happily disclose that I have an axe to grind. I think the major AI labs are an aggressive form of a cancer that's been ravaging our society. I want to see them fail, of course -- but more than that, I want to see the public develop an immune response to this.
I just can't wrap my head around why someone would expend so much effort speaking up on their behalf. They have, after all, highly compensated PR people doing that for them!
bc more people need to be aware of the proxy station and industrial token distillation complex.
many people i've replied to refuse to believe this is going on.
once you realize what's actually happening, and that you can get Chinese-lab-subsidized tokens at a >95% discount, why would you ever pay full price for overpriced APIs?
I suspect OP is the infamous Dario. We found his HN "anon" username.
That's the only explanation.
if Ford bought hundreds of millions of dollars worth of Hyundais, put extra instrumentation in them, and resold them at a discount to customers who agreed to the instrumentation in exchange for the discount, is Ford doing industrial espionage?
You skipped the part where Ford buys the cars at 90% off and sells them at 80% off, at a profit. Then gets paid by competitors for the driving data.
At the same time, Volvo is running the exact same hustle, except they buy the cars with stolen credit cards, so they get the cars for free.
You skipped the part where Hyundai chose to sell the cars at loss in the hopes of eventually gaining a monopoly position.
Misleading on both counts:
1) Anthropic tokens via subscription aren't sold at a loss, they're sold at cost.
2) Subscription plans are not sold in hopes of eventually gaining a monopoly position. They act as a loss leader designed to get a foot-in-the-door and funnel companies into costly enterprise plans, where Anthropic can charge full API rates.
Why should other people be responsible to make their business model work? They hold like all the money, if they can't make it work please shut down. Companies like OpenAI have already broken the public trust by breaking their non profit promises. They don't deserve any politeness at this point.
[dead]
Don’t sell your cars at a loss then.
Where are you getting this stolen credit card thing? That's so random.
Someone was claiming that it's not an issue for these proxy networks to create thousands of bot accounts and resell Claude's output because "they are buying the product" and "They pay for it!", which is a nonsensical position.
I responded that these resellers don't always acquire these accounts legitimately. They often use stolen credit cards, educational discounts, or resold compute credits to acquire them at essentially zero cost. They're not always paying customers.
That's one reason token resellers are able to price so cheaply, they acquire the goods for free.
Anthropic and OpenAI eat the loss.
> They often use stolen credit cards, educational discounts, or resold compute credits to acquire them at essentially zero cost. They're not always paying customers.
Yeah but where are you getting this from? I've seen this claim many places but only as pure speculation. No proof, just bold faced assertions.
https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens... https://x.com/yan5xu/status/2029743983522631698 https://x.com/howie_serious/status/2031620123413590471 https://x.com/Vincent_AINotes/status/2046434813125763527
The china talk page points to an article about crypto currency stolen credit cards. If I believed every random X account i would have to believe too many false things. Not really convinced my guy.
The companies running these proxy stations are already:
1) Using botnets to mass-create thousands of accounts
2) Blatantly violating ToS by splitting and reselling accounts
3) Creating thousands of accounts using fraudulent identities
4) Bypassing KYC by recruiting real people in low-income countries for biometric face-matching checks for a few dollars
5) Using AI deepfakes to fake passports / verification credentials
But if you believe that payments fraud is the one ethical line these syndicates refuse to cross, there's not much else I can say to convince you.
assuming the k3 model weights do indeed get published, if your model of the world is "achieving RSI is beneficial and K3 has done so," this feels structurally different from ordinary industrial espionage, because the knowledge has enriched the commons
more like silk than capacitors
if, again, your model is that RSI will be beneficial, why wouldn't making it available to all unlock more benefit globally than not doing that
>through proxies and heavily discounted token resellers
Could you explain a little more about how this works? Are you saying that the Chinese run or have backdoored something like OpenRouter?
Have a look at https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens... and https://x.com/yan5xu/status/2029743983522631698
Chinese resellers acquire hundreds of Claude Max 5x accounts and set up a custom proxy server. Customers point their ANTHROPIC_API_KEY at that proxy, and requests are routed to Anthropic through one of those hundreds of accounts. Because one $200 Claude Max 5x account gets the equivalent of ~$2000 in of API credits, these resellers can resell Anthropic tokens at a massive discount, undercutting official API prices by more than 90%.
To cut costs even further, these accounts are funded using educational discounts, startup credits, or stolen credit cards.
The resellers log all data traveling through their proxy networks, which they then resell to Chinese labs as high-quality training data for significant profit. https://x.com/xkajon/status/2050445443889525235
The resellers also loan these proxy networks to Chinese labs, allowing them to can run distillation attacks on Anthropic, while blending in with regular user traffic. https://www.anthropic.com/news/detecting-and-preventing-dist...
This is a widespread tactic, there's hundreds of proxy resellers operating. Some even offer enterprise SLAs.
oh wow, an entire seedy underbelly I was unaware of. Thanks, great reply. Appreciated!
> nobody cares at all that it happened.
Who in their right mind would care? Why care? Misplaced patriotism?
"A thief who steals from a thief has 100 years of forgiveness". Spanish proverb.
In fact, I would be very concerned about the sanity of someone who cared about this sort of thing, unless they were Dario themselves.
Why should anyone care? I couldn't give a single fuck, in fact if what you assert is true (definitely not proven), I applaud Moonshot - seems like a very smart way to operate.
> nobody cares at all that it happened
Oh, no. I wouldn’t say that. If that happened, I definitely care: I’m positively delighted about it.
They stole from me first. And are spitting in my face and telling me they’ll take my job while they do it. I have negative sympathy for them.
[dead]
But then again, the identity could also have slipped into the model from other sources during pretraining. The internet is full of "I am Claude": https://grep.app/search?q=i+am+claude and variants https://grep.app/search?q=i%27m+claude
Either way, there's probably no significant portion of Mythos/Fable or Sol in there as OP has stated.
fwiw, Gemini 3.5 has identified itself to me as an OpenAI product on multiple occasions.
Early Grok would also identify as ChatGPT. This has happened with new model releases for years now.
Claude Sonnet 4.8 reproducibly identifies itself as DeepSeek when asked in Chinese:
https://x.com/stevibe/status/2026227392076018101
I mean, people can point fingers however they want, and the fact is nobody actually "owns" the data they feed to their LLMs...
The entire debate about distilled vs not distilled is just academic. As an end user, I don’t give a crap how the model was trained as long as it does the job affordably enough
I see all of AI as theft anyway so it makes no ontological difference if the theft was from a human or from another AI
If your business leaks to your customers… if you can ask your product for its inner secret sauce and it readily gives away the goose. That is not a moat.
The desire to accuse China of just copying is like 20 years out of date. It’s been wrong since some people on HN were in diapers.
People are going to be gobsmacked when, in our lifetime, China becomes a world power comparable to the U.S. Probably still poorer per capita, but at Spain/Italy levels, not third world country levels. And they’ll be shocked at the implications of that on the world economy, migration patterns, etc. There will be fields where China is a global leader, and Americans and Europeans will have to learn Chinese and move there, or else be stuck in some satellite office of a Chinese company. We’re all in Europe circa 1895 not realizing the behemoth America will become in WWI.
I am still shocked Spain/Italy and USA are considered 'first world' countries. We are not in 70s or even 90s anymore. I've been to China in 2011 also thinking I am visiting some huge village but...that was the most futuristic trip I ever had. I was surprised by the penetration level of the mobile devices - everything had a QR code, you could buy/sell/send money, pay services all with a single tap on a phone.
China has pursued a phased approach to growth. Instead of trying to pull everyone up at once, it’s trying to get top tier cities to a high level of development before moving to other cities. Shanghai already has a GDP per capita (PPP) that’s comparable to Madrid. But there are provincial capitals with almost 10 million people that are less than half or less than a quarter of that.
Visiting cities is also a misleading way to compare the U.S. in particular to anywhere else. I have family in town in Mississippi that has less than 10,000 people. But the town has a household income over 60% of the national household income. Cost of living adjusted, they’re about as well off as someone in a top tier city. Someone with a median household income can afford a newly renovated, 4 bedroom, 2,500 square foot house.
I've been in China in 2015 and like anywhere else in the world it was very mixed: some urban areas like central NY or central Madrid and Milan (or much shinier) and some rural areas like 200 year ago, but inevitably with electronics.
Basically in every country of the world you can travel one hour from big cities and get in a place deep in the fields or the woods with very different needs and dynamics from the city. They could be different countries and maybe both cities and countryside will be better off if we could have fractally composed states with different laws and regulations.
I don’t know any cities in Americas that is comparable to Shanghai or even Japan in term of transportation or convenient
Americans optimize for hyper-individualistic convenience. The average one-way commute in Dallas, Texas is under 30 minutes. The average in Tokyo is 48-50 minutes. And the guy in Dallas goes to work in a perfectly climate controlled bubble where he doesn’t have to interact with anyone else, while the person in Tokyo is crammed into a rush hour subway car.
I love Tokyo too (never been to Shanghai), because I’m an asian collectivist at heart. But you can’t really compare across cultures when they’re optimizing for different things. Americans are very wealthy and spend a lot of their wealth optimizing to never have to be near other people.
Paris has a metro station everywhere at least in what a tourist can assume to be an enlarged city center. Tokyo is another city with a lot of metro stations. Manhattan too, at least up to Central Park (but 20+ since my last visit.) I don't remember Shanghai to stand out positively or negatively, but 11 years can be a long time.
I mean, you have those kind of luxuries even in the poorest of countries in Asia, it’s just that there’s still a huge discrepancy between rich and poor, city vs countryside.
It’s not difficult to find areas in all these countries that are significantly less developed than Spain/Portugal’s underdeveloped areas. It’s just not as black and white as you seem to suggest.
(I come from EU but have been living in various countries in Asia for over a decade)
Not to disagree with you too much, but
> I am still shocked Spain/Italy and USA are considered 'first world' countries.
They're a mix. Rural southern Italy isn't the same as e.g. Milan or Venice. I've walked from 1st world to third world within a few blocks in San Francisco. It's a slightly longer walk in Cape Town.
> I was surprised by the penetration level of the mobile devices - everything had a QR code, you could buy/sell/send money,
I've has exact same experience in places in Africa (1). Yes there's poverty and crime, but also if the technology is affordable, effective and reduces the need to handle cash then it's adopted fast enough.
People's understanding of that part of the world is also decades out of date. Mobile devices actually "leapfrogged" the wired telecoms network rollout (2), but that was decades ago. Africa is huge and diverse, and it is not going to be China this decade, but also it's changing fast.
And it might be China-aligned as China positions to be a reliable trading partner with affordable goods. It's possible that affordable Chinese solar-battery electricity systems will cause another leapfrog. This includes Chinese EVs (3).
1)
https://www.m-pesa.africa/
https://www.payshap.co.za
2) https://mg.co.za/news/tech/2014-06-12-cellphones-create-a-te...
3) https://www.youtube.com/watch?v=Fo---4TAIEA&t=528s
There are huge evidence of copying.
Some day China can pioneer in science or technology but the current claim about Chinese companies leading AI development is ridiculous given the evidence of distillation and the fact that like 95 percent of science that lead to the current state of AI happened in either North America or Europe.
To be honest if you want to list academic papers that lead to the current AI models the majority is either done by Google Research or sponsored by Google.
In 2017 maybe. This chart shows last year’s Neurips accepted papers by country and institution (top 50). What is missing here is that the papers from American institutions also have mostly Chinese authors. Europe is sliding and Singapore has more papers than Canada.
There is a clear trend.
https://www.reddit.com/r/accelerate/comments/1pi64q0/papers_...
US is still winning because of their hardware dominance. Also they have astronomical budgets and much better financing. They throw money at an industry until they win. Whereas China throws lots of (educated) people at it. 38% of top AI researchers today have Chinese education and origin^. And hardware dominance will change in the upcoming years.
^ https://archivemacropolo.org/interactive/digital-projects/th...
Volume is up because AI generation help writing papers. We should find a better measure of impact. I like to see the same charts on best paper awards.
From the Hoover Institution’s analysis of the team behind DeepSeek:
“We find striking evidence that China has developed a robust pipeline of homegrown talent. Nearly all of the researchers behind DeepSeek’s five papers were educated or trained in China. More than half of them never left China for schooling or work, demonstrating the country’s growing capacity to develop world-class AI talent through an entirely domestic pipeline. And while nearly a quarter of DeepSeek researchers gained some experience at US institutions during their careers, most returned to China, creating a one-way knowledge transfer that benefits China’s AI ecosystem.”
That was from a year ago.
Consider that on top of this the country was starved of access to Nvidia chips - and therefore accelerated its development of Ascend chips, and it’s clear they are undeniably leaders in AI research and development. Not the only ones, but the achievements are crystal clear.
Exactly. China is a real tech power now, just like Japan and Taiwan. The U.S. is ahead in a lot of areas of technology, but China has home grown talent that is taking the lead in other areas. And unlike Japan and Taiwan, China has a much bigger pool to draw from.
Okay but I cannot stress this enough: no one cares.
It's international politics. The rules are optional, and written on the back of whoever agrees to enforce them.
If you're going to run around declaring AI is a strategic advantage vital to national security, then guess what? Stealing it is a great idea. That you stole it is only a problem if it means you're not developing the ability to support that work locally as well, and China seems to be doing very well at building it's local talent and support network.
If you ever listen to Russian propaganda, there's a similar theme: every big idea, everything good, all of it was definitely first developed in Russia - only Russians could ever have thought of it. Of course, Russia isn't actually a world leader in any of those things, or able to execute on them.
Which is what America is sounding like more and more these days.
> If you ever listen to Russian propaganda, there's a similar theme: every big idea, everything good, all of it was definitely first developed in Russia - only Russians could ever have thought of it. Of course, Russia isn't actually a world leader in any of those things, or able to execute on them.
When I was a kid watching Star Trek VI, I was confused by the line "You've not experienced Shakespeare until you've read him in the original Klingon".
And then I learned about how the Klingons (especially in that film) were a stand-in for the USSR.
But that was more of a jab at literary snobs who would tout "Homer in the original Greek" or "Marcus Aurelius in the original Latin" or "Old Testament in the Original Hebrew". It has been such a meme, probably for centuries. Because it was not so long ago when university students were actually conversant in many classical languages such as those.
What about EVs, solar, or batteries? They are leading these fields for some time.
I mean, China is the leader in PV panels, battery tech, telecommunication hardware, etc... already...
[dead]
[dead]
So the efficient market hypothesis is wrong?
This isn’t even controversial assuming you’re talking about the real world, economists freely admit that. It only holds for spherical markets in a vacuum.
What do you mean, I don't follow.
Also, yes, often.
How is the efficient market hypothesis applicable here?
Almost all markets depend on some form of regulation whether its as simple as "leave everyone alone but no stealing" or "every participant has to source every object through mountains of red tape."
Thus far the US has not really chosen to go the Chinese rare-earth method yet. The problem with distillation attacks is the end result is everyone who is not doing them is going to deal with some kind of regulation whether it's complete loss of access, or the amount of control you'll have to give up to access them will be ridiculous.
Sort of like the "stealing music is fine" but "lets freak out now that it's producing visual art", in the end the entire thing is a social construct. Whether this is treated as theft or "business as usual" is entirely societal.
Eventually the gap will close, unless there's a major breakthrough that hasn't been made yet.
Given these models could not have been trained in the first place if they had to license every line of random fan fiction on the internet, I think distillation also being fair game is a tradeoff everyone should be willing to take (unless they want to decelerate, but that's a different conversation).
Us models didnt pay for licenses too
We're still in the early days of the AI industry timeline(relative to traditional industries). Not everything has yet been litigated.
Taxes on AI subscriptions or AI capable hardware, to financially compensate IP holders for (potential) IP theft, could very well arrive in the near future, once the industry is mature.
If this shocks you and sounds preposterous, I'll remind you that in several EU countries, we still pay extra taxes on any and all storage mediums and on devices with built-in storage (tapes, CDs, DVDs, HDDs, SSDs, tablets, phones, etc) simply because they can be used to store pirated content, decisions based on laws from 50-100 years ago, and the money goes to the national unions and associations of music and arts IP holders. It's basically a lobby pushed and government legalized extortion racket that no voter agrees with or can change but has no choice but to conform either way.
So I guarantee you in the future, it will be the same for AI subscriptions and hardware capable of running LLMs locally. Every time you purchase a Claude or ChatGPT subscription, an Nvidia GPU, Intel/AMD SoC PC or an Apple/Qualcomm powered smartphone, you'll pay a government enforced tax to the likes of Sony, Axel Springer, etc. for licensing their IP, whether you want to or not. In the EU at least. US maybe not.
I think we are going to direction where AI corps will have stronger lobby compared to IP holders.
giving peanuts to the other guys is a very well trodden strategy to keep in power tho
[dead]
That is incorrect. Anthropic paid $1.5 billion in compensation to copyright holders for use of their content in training data. OpenAI pays hundreds of millions per year across 150+ licensing deals for access to copyrighted data. Meta and Alphabet have similar arrangements.
Under the settlement, Anthropic was forced to delete the pirated data they were training on.
Chinese labs can still train on pirated data. I doubt the Chinese models operate under similar licensing agreements.
Anthropic paid $1.5 billion in compensation to copyright holders for use of their content in training data.
The payment was for illegally downloading copyrighted material, not training. Training was explicitly ruled to be fair use.
Partially correct. The court explicitly ruled that training on pirated data, which is what Anthropic was doing, is not considered fair use.
Training on legally acquired / licensed data is potentially fair use.
It's not potentially, it's settled. At least for now as neither case wanted to move on to appeals
Not at all. The ruling came from a federal district court, and since it was settled early, it was never reviewed by a higher court. It doesn't set a national precedent across the U.S.
And other district courts don't agree on this. The US district court for Delaware recently rejected a fair use defense for the use of copyrighted works to train AI. https://www.reedsmith.com/articles/court-ai-fair-use-thomson...
There are more cases in the pipeline. The massive NYT vs OpenAI is still ongoing. Nothing will be "settled" until this makes its way to the Supreme Court or Congress steps in.
they didn't pay yet, because court challenged settlement as inadequate.
> I doubt the Chinese models operate under similar licensing agreements.
US corps likely pay licenses when afraid to be sued, or have troubles getting that data, otherwise they just take data, which was demonstrated many times. The same apply to Chinese corps, alibaba totally can be sued in US.
China is infamous for weakly enforcing copyright law. Even when it is completely obvious that Chinese labs are training models on pirated data, US copyright holders face a virtually impossible task of proving it in court. Those lawsuits won't go anywhere.
The US is currently infamous for weakly enforcing copyright law when it comes to AI companies.
There are tons of lawsuites which resulted in banning Chinese companies from doing business in US, those lawsuits totally have consequences.
What are the most high-profile examples of the "tons" of lawsuits resulting in Chinese companies being banned from doing business in the U.S.? Isn’t it usually more action by the government - executive orders, etc?
Here is example: https://www.scmp.com/tech/tech-trends/article/3258239/chines...
I believe mechanics is following: US corp sues Chinese, asks for preliminary injunction to stop selling product for example if there is strong evidence some IP for example was stolen etc. Then they litigate, and settle somehow.
That 2024 article says "US sanctions" in the first sentence, but it's paywalled, but https://en.wikipedia.org/wiki/Hytera#United_States first mentions a 2019 US law that first partially banned them, with the US government subsequently expanding it to a general US ban. After the initial ban it appears Hytera was involved in a suit with Motorola and got a worldwide(!?) ban as a result of it in 2024, but the ban was lifted on appeal after 2 weeks (just after the SCMP article). So it appears Hytera was first banned by US law, then got a 2-week worldwide ban from a US suit. (I'm just relying on the linked sources and have no personal knowledge of all of this.)
Sure, there is litigation, criminal case, appeals, fines ($500M: https://www.motorolasolutions.com/newsroom/press-releases/hy...). The point is if violation is clear, US corps have a chance to go after Chinese corps.
>> There are tons of lawsuites which resulted in banning Chinese companies from doing business in US
> What are the most high-profile examples of the "tons" of lawsuits resulting in Chinese companies being banned from doing business in the U.S.? Isn’t it usually more action by the government - executive orders, etc?
In response to "What are the most high-profile examples of lawsuits resulting in Chinese companies being banned from doing business in the U.S.", the one example given was from 2 years ago of a ban that lasted for 2 weeks (separate from its 2019 onward government bans)?
However, if the claim is that companies (including Chinese) can face significant fines from IP lawsuits, I agree.
They settled with a subset of copyright holders. Guarantee they violated lots of others' rights in the process
They only paid when they got caught. And not to everyone.
But they still paid. I don't see any Chinese labs paying billion dollar infringement settlements.
Chinese labs can freely train on pirated material, which is a structural advantage.
really!? nobody paid me anything for my comments on HN.
The only ones getting paid this time around had registered copyrights (in the US at that.)
Compensation is not license
Let’s not forget that Anthropic only paid that to settle a class action lawsuit.
They used two of my books and I'm still waiting for my cheque here.
That's like saying someone is a big proponent of community law and order, and they donated $1000 to the county sheriff when actually they got caught drunk speeding in a school zone.
A false equivalence. A more correct example is: Anthropic was speeding, got caught by the county sheriff, and paid the fine. Anthropic stopped speeding.
Meanwhile, Chinese labs are speeding in a different county. Everyone knows they are speeding, yet the sheriff won't pull them over, so they just keep doing it.
This lax enforcement gives Chinese labs a structural advantage over American ones.
> Anthropic stopped speeding.
Do you purport to know for a fact that they're no longer training on the data they'd pirated? Because I highly doubt that.
Anthropic deleted the pirated training data as part of the settlement https://www.ropesgray.com/en/insights/alerts/2025/09/anthrop...
Destruction of Materials: In addition to the monetary compensation, Anthropic has agreed to destroy the two libraries that allegedly contain the pirated works, as well as any derivative copies originating from those sources. Anthropic must certify in writing to class counsel that the destruction has been completed and that the allegedly infringing materials are permanently removed from its systems.
The libraries in question were Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi).
If Anthropic is somehow training models on deleted data, I'd be quite impressed.
Because they got caught
there is much less intellectual property in China so it’s not ‘theft’ (as you can’t put property on information)
After the fact. They did the same thing Youtube, Uber and Airbnb did: Break the law, eventually get caught, cut some deal where they pay a pittance and keep doing the same thing but now with leverage on their side.
How is distillation an "attack" but gigascraping the Internet to the point of crashing servers and everyone needs Cloudflare and Anubis now not an "attack"?
I'm not aiming for a what about kickflip here: I'm saying we need to either agree on some rules or stop crying foul. Maybe the coherent legal theory is that neural networks and intellectual property don't interact. That would be weird but it would be consistent, a market could price it, I could do coding stuff and know if I was illegaling.
But this weird gerrymander that no judge will really rule on in an emphatic way is like, bad for the planet, bad for markets, bad business.
There are a lot of reasons to look forward to DeepSeek Huggingface drop kicking the unambiguous frontier weights in like, November, but I think my favorite one will be "who's distilling now bitch?"
I think you've basically got the legal theory. Training a neural network isn't prohibited by copyright law so if you can legally get your hands on something (e.g. by sending a GET request to someone with rights to serve the contents of their web page, or by buying a book) without signing a contract to not train on it, you can train on it.
But the American AI companies only let you query their models if you first sign a contract to not train on the output.
It's hypocrisy and unfair, but I think there's a strong legal argument for it.
Of course China can simply decline to assist in enforcing that contract... But I would expect US courts to do their best to.
> to someone with rights to serve the contents
Now THAT'S doing some heavy lifting lmao. The vast, vast, VAST majority of the original datasets were from pirated books and the like. Also, arguably a robots.txt is the exact mechanism to follow to do the mass GET-ing, yet the AI cos choose time and time and time again to simply ignore it and be as abusive as they possibly fucking can
> The vast, vast, VAST majority of the original datasets were from pirated books and the like
And there's been significant legal consequences as a result
> Also, arguably a robots.txt is the exact mechanism to follow to do the mass GET-ing
You're free to argue this of course, but the courts have largely rejected it already pre LLMs. See for example hiQ Labs v. LinkedIn
Yes, anthropic and openai have really been brought to their knees and ipos cancelled because of the legal consequences of obtaining their training data.
This would have been a problem but it turns out that Anthropic is actually valued multiple orders of magnitude more than a copy of all the books in the world. So they survived the significant legal consequences.
In a just world, the punishment can be more than just the sum of the direct damages, otherwise there's no incentive to stop reoffending.
Anthropic (and friends) proved they're willing to do obviously illegal things, and it didn't end them. Why do we think they stopped after doing it once?
In this world too, the punishment was, ballpark, 100x direct damages.
Contract law is never going to prevent this.
Why not? Seems like a perfectly normal contact term to me.
Or do you just mean that US courts don't have enough teeth to prevent Chinese companies from violating contracts? On that I agree.
I mean the latter, but more narrowly: China would never allow the United States to have a monopoly on machine intelligence if the only thing standing in the way of a domestic alternative was the Anthropic ToS. In general, I think that China is willing to agree on certain things relating to intellectual property. But not on this, it’s too big.
The US is already publicizing the way they are using Claude with Palantir for war gaming purposes. It’s a matter of national defense. Contract law has no meaning here.
> It's hypocrisy and unfair, but I think there's a strong legal argument for it.
That right there is the problem.
> but I think there's a strong legal argument for it.
Maybe today. I doubt it tomorrow. Legal and not legal, largely, has to answer to the population sooner or later. Ultimately, humanity decides legality. And I don't think the frontier labs will get a pass from humanity in the midterm, let alone the long term. I think you'll see the rules change towards something more "intent" driven. And there's absolutely no difference in intent between Frontier labs and everyone chasing them.
Frontier labs just want the door closed behind them, as do their investors, because they know the money will never be recouped if others can do the same magic tricks.
Eh, I think you've done a pretty good job summarizing a collection of settlements with a few narrow bench rulings for seasoning. I'm not sure I follow you to it being a coherent legal theory. Buying a book in a bookstore is sure legal, and excerpting from it for e.g. literary criticism is pretty settled. Downloading every torrent of all e-books ever is pretty clearly illegal (or at least it fuckin would be if I did it). Pretty sure like, multiple labs have been popped for that though.
Situation right now seems more like a fragile detente: if you got a Hill staffer drunk and hounded him long enough he'd probably be like "God damnit the market will fucking tank if we don't get these two IPOs out north of a trillion. And don't even get me started on how I'm going to sell Chinese AI to a Senate that still calls people Nipponesians when no one is looking. We're doing the best we can alright, get off my back man."
We have a situation, but it's not exactly A&M Records, Inc. v. Napster.
> Downloading every torrent of all e-books ever is pretty clearly illegal (or at least it fuckin would be if I did it). Pretty sure like, multiple labs have been popped for that though.
Oh it is, and at least anthropic has paid $1.5 billion and deleted there torrented copies and not released any models derived from them as a consequence.
The thing is it turns out to be not that expensive to just buy a copy of every book legally and scan them. And there's even precedent that this is legal predating LLMs (Google books)
> and deleted there torrented copies and not released any models derived from them as a consequence.
I have a bridge to sell you
Great, let's go down to the courthouse and get some sworn testimony as to the ownership, value, condition, and so on and so forth of the bridge. And some document review and discovery run through professional legal firms under the same conditions. And perfectly reasonable and verifiable explanations as to why you own the bridge and are selling it (namely that you bought a copy of literally every book in existence in the meantime).
Facts are in fact knowable, and the US legal system is in fact not terrible at getting to them.
I think you're right to point out that historically the rule of law in the United States has been very robust by the standards of whatever era, it's been a tremendous advantage in attracting business and capital and talent, it's good stuff.
But we've gone through some pretty weird times too. Turn of the last century was pretty tech billionaire edits, reconstruction was uh, not smooth, it's a mixed bag.
And most takes I hear seem to acknowledge that this is one of those weirder times: serious election fraud rhetoric from most everybody from 2016 to the present, very politicized courts (on both sides to be clear), very soft on anti-trust, very soft on adventurous accounting. The Epstein files and like, no consequences (pretty much uniquely for a developed nation with Epstein people). It's weird right now.
And I think I would be hard pressed to think of a weirder part of this weird time than the rule of law meets AI. We can haggle on where laws end and norms begin (stare decis being maybe the midpoint), but in the 90s, the Justice Department got their brass knuckles on for a lot less.
I don't think it's a simple "the law works nothing to see here" story.
I broadly agree with your take on the state of the US - but this is a case where given the specific facts at hand I'm confident it still got to the truth.
I can understand why as someone who didn't follow it and the more corrupt legal developments closely you wouldn't be confident in that.
Knowlege should not have ownership. Training and distillation should be allowed
Granting people some form of control over knowledge only serves the public interest inasmuch it provides incentive to create more of it. Mass media, effortless duplication, and copyright extensions had already broken this to the point where control of knowledge was suppressing creation of new knowledge more than it facilitated.
The world has changed, we need a mechanism that works for the public interest that applies to the facts as they now are.
American labs have ripped everything out of the internet. And now they cry someone else is “stealing” from them. Cry me a river.
Apologies for repeating myself here, but what you call "distillation" is function approximation.
I feel for the teams at Anthropic and Open AI, but unlike startups from prior eras; Anthropic and OpenAI have decided to be in the business of selling compute. Not creating a product that uses compute, but a product that's math running on compute. This is different from what Google is (or, rather was. As always, RIP Google 1998-2019).
Google's algorithm might be math, but Google search isn't. Google search is a process that's continuously operating in the background. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Now, let's compare that to AI models. When Anthropic serves Mythos / Opus etc, they're taking input or x from their user, doing compute, and then serving the result of the Mythos / Opus function, i.e.,
Where f is a continuous function, https://www.turing.ac.uk/sites/default/files/2025-11/languag...According to Stone-Weierstrass, given enough values of y for f(x), anyone can approximate this function.
The fidelity and sophistication of this approximation definitely requires a lot of cleverness and effort, and it is arguably an imposition on Anthropic and OpenAI. But on a long-enough timeline, they don't even have to poll Anthropic or OpenAI. As the internet is flooded by PRs, content, emails written by Mythos / Claude, and just people otherwise sharing the results of Claude prompts, then there's an ever increasing set of data to approximate the f(x) that's f_Claude.
Eventually, in the future, anyone will be able to create a good enough approximation of the f_Mythos. Which is Anthropic's product.
Anthropic and OpenAI can now wage war on mathematics and the open-ended compute. Or, they can adapt and build a better product.
Choosing Option B was the Silicon Valley option / choice. I think the OG large-scale Valley lobbying effort, the Semiconductor Industry Association, was unique in that it prioritized and chose to do real research.
https://en.wikipedia.org/wiki/Semiconductor_Industry_Associa...
https://en.wikipedia.org/wiki/Semiconductor_Research_Corpora...
This helped the industry to survive and outcompete the pressure they were facing (at the time).
I like your point that there is so much content being created by LLMs that at some point there’s enough to perform something like distillation without even needing to interact with the LLMs directly.
This is nothing like music piracy.
The court decided that LLMs are a transformative fair use of the data they trained on, and therefore aren’t copyright infringement.
Maybe Kimi is a derivative work as well
LLM outputs do not have copyright protection in the US, there is no copyright element here.
It is unfair, they stole the dataset that we stole.
Calling distillation an 'attack' is exactly what I've been describing as "AI Exceptionalism":
https://www.magiclasso.co/insights/ai-exceptionalism/
Look how hard Anthropic is to even be able scroll back on your conversation, or look at the thinking tokens or subagents. They want to keep everyone coming back to the watering hole but never to learn how to dig a well.
Did you enable the flicker-free TUI mode?
Why is it hard to scroll?
It really is not, not sure what OP is on about.
On some terminals it is.
A “distillation attack” is like a concrete company calling a competitor building a factory with its concrete a “construction attack”
I wonder, how does distillation deal with unprobed spaces in the knowledge landscape? Is a distilled model worse in some niche area that was not probed? Presumably, this is why frontier labs dont distill their own models internally to release them to the public as a servicable frontier model.
Do frontier labs not distill their own bigger models into the smaller/cheaper variants? I thought that’s been the case for a while
if distillation is the key, why the fuck all other competitors do not release competitive models? and only Chinese can distill this great?! Am I smoking too much?
are there any "open source" efforts to do distillation? Like some place one can submit one's anonymized chat logs? So they can be pooled and used as an open training set (similar to OpenCrawl)
Pulling on this thread, if the model companies become commoditized and make no money then who is buying the hardware? Seems like it would be the next shoe to drop
I suspect that distillation attacks may be slightly exaggerated. Most of the training data used during fine-tuning is now synthetic data. You can't just repeat the same stuff twice, therefore another LLM is writing a text book that is explaining a topic in detail, ideally without any gaps in the material.
The fact that API based distillation is even a conversation right now makes me feel like the U.S. has their heads so far in the sand that it’s not really excusable.
These Chinese labs are producing novel models, publishing their techniques and sharing their open weights and the first topic of conversation is how they stole from U.S. AI labs.
Setting aside the fact that it doesn’t make any feasible sense to do API distillation, these models are outperforming frontier models on a number of benchmarks, and often times run more efficiently by several orders of magnitude.
We have to stop crying distillation, it’s getting embarrassing and at this point feels even a bit delusional.
> We have to stop crying distillation, it’s getting embarrassing and at this point feels even a bit delusional.
It's a PR campaign - when they say its an "attack" they don't mean on Anthropic - but on America itself. What kind of American can let such a brazen attack go unanswered? At the very least, they ought to demand the dangerous, pinko, stolen models be banned in all 50 states, and pay whatever price demanded by the patriotic, freedom-loving, all-American AI labs that can never be accused of stealing.
[dead]
There's little doubt that Kimi K3 was distilled off Claude.
Anthropic stated in February that Moonshot AI (the creator of Kimi) distilled ~3.4 million exchanges from Claude models, as explained in their press release https://www.anthropic.com/news/detecting-and-preventing-dist...
It’s so funny to me that Anthropic can make claims like this one with zero evidence provided.
DeepSeek and others like Minimax are publishing deep research on Multi-Head Latent Attention and Mixture of Experts, Multi-Token Prediction, novel Sparse Attention approaches, I mean they trained long context models on a fraction of the resources and gave everyone the recipe.
Chinese labs might not have the funding of labs like Anthropic, but at least they provide the receipts.
There's reproducible evidence of Kimi K3 spontaneously identifying itself as Claude https://x.com/denisewu/status/2077984660211269870
This behavior is exactly what you'd expect from a model distilled from Claude.
Someone even took the time to analyze Kimi's ambiguous identity, in great detail: https://github.com/rgreenblatt/which_claude_is_k3/blob/main/...
And there's an entire Reddit thread discussing this https://www.reddit.com/r/LocalLLaMA/comments/1m2w5ge/did_kim...
That doesn’t prove Anthropic’s specific 3.4m-session allegation, but calling it “zero evidence” is no longer credible.
Kimi K2.5 was worse in a hilarious way, it identified itself as Claude and referenced Anthropic's Constitutional AI as some of its guiding principles https://huggingface.co/moonshotai/Kimi-K2.5/discussions/38
> This behavior is exactly what you'd expect from a model distilled from Claude.
This is not at all what I would expect because it's trivial to change the training data to replace Claude with Kimi. In fact I'd argue it's almost certainly not saying that due to distillation.
I encourage you to review the links before committing to a position. The writeup on K3's anomalous trans-model identity is very comprehensive.
K3 reproduces Claude's internal model identifier when prompted, something which the real Claude models themselves do not emit. This is highly suggestive that K3 was trained on Claude metadata (API logs, tagged synthetic data), rather than Claude's chat outputs.
And it's well documented that Chinese labs are buying large amounts of raw Claude metadata https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
For context you just asked me to read a document that starts with:
"Caveat: fully AI-generated research."
And that you quoted or paraphrased directly.
>This is not at all what I would expect because it's trivial to change the training data to replace Claude with Kimi.
Wait what? The reason you wouldn't expect it is because if it was distilled, it would be easy to get rid of self identification? Is that any less true of a non distilled model? I suppose there's lots of ways to interpret it, but the idea that self-identifying as Claude is affirmative evidence that it's not distilled seems to get the weight of the inference exactly backwards.
By evidence I mean logs, I mean IP addresses, I mean timestamps. They claim millions of requests, let’s see literally any of them?
I don’t consider a tweet by Denise Wu, who works at Anthropic, to be reproducible evidence.
I don’t consider “Caveat: fully AI-generated research.” To be someone taking time to analyze anything in great detail.
Because two AI models produce vaguely similar front-end styles when generating similar prompts I also do not consider to be of much value?
I think this is what I mean when I say the U.S. has its head in the sand. The Chinese labs are releasing ~60 page research reports with citations and analyses and evidence and Anthropic is throwing up defensive blog posts with zilch. I’ve seen more detail in a tech blog from Uber than anything I’ve seen from Anthropic.
You've backtracked significantly here.
"Zero evidence" as you claimed earlier isn't accurate. You've moved the goalposts from "evidence" to "raw internal logs I can independently audit," which is a different and very high standard. Sure Anthropic didn't publish logs, IP addresses, timestamps, or account IDs of the accounts involved. But that's true of any cybersecurity breach/abuse disclosure ever made. Companies are furtive to reveal how they detect fraud, because doing so exposes the signals used to detect bad actors, and makes future abuse easier. Not revealing the "evidence" you're asking for is industry standard practice. You're complaining that Anthropic is following industry standard practice, and conveniently defining the "evidence" you need as something Anthropic is never going to publish.
> I don’t consider a tweet by Denise Wu, who works at Anthropic, to be reproducible evidence.
Is the issue here that she works at Anthropic? Because Denise Wu doesn't work there.
> I don’t consider “Caveat: fully AI-generated research” to be someone taking time to analyze anything in great detail.
The experiments were run by Ryan Greenblatt, who is a real AI safety researcher (at Redwood Research).
The identity experiments and Greenblatt analysis are trivially reproducible. The methodology, code, and metrics are all there in the Github repository. You can ask your preferred AI to independently replicate these results, and it will give you a result within an hour.
You’ve also reduced the evidence to “two models producing vaguely similar front-end styles,” which is not what either analysis shows.
From the analysis, Kimi K3 identifies itself as Claude 15% of the time. How do you explain that? Qwen and GPT identify themselves as Claude 0% of the time.
If a long document is too much analysis for you, someone else made a simple chart which measures the KL divergence between Kimi K3 and other major models. They found K3 is unusually similar to Fable 5 & Opus models. That is, Kimi K3 has an very similar style and phrasing to that of Anthropic models. That behavior is expected from a model distilled from Claude.
https://typebulb.com/u/lab/you-re-relatively-right/full
"From the analysis, Kimi K3 identifies itself as Claude 15% of the time. How do you explain that? Qwen and GPT identify themselves as Claude 0% of the time."
Qwen and GPT have special guards that trigger when asked to identify, Kimi doesnt. I dont understand the argument. Kimi is an LLM and does not know what it is. It will give you the most likely answer which sometimes is Claude.
I wouldn’t say I’ve backtracked- I think I’ve been incredibly consistent here. Chinese labs are releasing open weight models, research and analysis. Anthropic is not. They haven’t produced any actual evidence of distillation themselves and what they have presented is tenuous at best.
While it sounds like a lot, do you suppose 3.4 million sessions come even close to being sufficient to train a frontier model?
Assuming each session was 10,000 words each, that's 34 billion words; lets call it 50 billion tokens (0.05 trillion) unfairly pilfered from Claude. That left Moonshot needing to scrounge for the other 14.950 trillion training tokens required for a baseline frontier model.
What do you think those tokens are used for?
Distillation attacks aren't about replacing the entire pretraining dataset with questionably sourced synthetics. It's all about post-training.
Train your own base model - but tune it off Claude output to make it perform more in line with Claude. Yoink the products of Anthropic's expensive SFT, RLHF and RLVR work for yourself by training on the outcomes.
The post-training datasets are small, but they are what controls the final model behavior.
> Train your own base model - but tune it off Claude output to make it perform more in line with Claude
Is that actually genuine distillation though? Distillation suggests the core model is being pre-trained using output from another model. For the above to work, you have to already have all the core intelligence trained into your base model.
If distillation just comes down to post-training then it's tantamount to admitting that the Chinese base models are just as good as frontier US lab models. Because you can't post-train frontier intelligence into a model. It has to be there in the base. Then you can change how that intelligence is expressed through post-training.
What's in the base model is "bits and pieces of intelligence".
You have to bring those bits and pieces together, put them into the right shapes and fill in the gaps to get a model that actually performs. This is what post-training is all about. It's not at all a trivial thing.
Reasoning, tool use, agentic behavior - all of those are post-training performance gains. Getting a good well trained base model is putting your foot in the door of frontier performance - post-training is how you actually get inside.
See: GPT-4.5 vs o1. One went for "build a bigger better more capable base model", the other went for "take the old base and post-train it for advanced capabilities". The results: a wider base with basic post-training loses to a narrower base with advanced post-training. Or, hell: GPT-3 vs GPT-3.5. One was largely a research lab curio, and the other kicked off the AI revolution as we know it.
The gains compound. Getting a better base model with the same type of post-training helps, see: the jump from Opus to Mythos/Fable. But post-training techniques account for a lot of the performance juice.
And yes, reasoning trace post-training distillation is "genuine distillation". As is logit distillation in pre-training. "Distillation" isn't a single training recipe that you have to follow to a tee - it's a large group of training methods. I've seen plenty of wacky things like inverse distillation bootstrap and post-training self-distillation that use distillation in strange ways at different stages of the training run to get results.
How does yoinking outputs from from prior generation Claude model and post raining on them result in a model competitive with the latest generation? That doesn't add up - nevermind Anthropic hasbeen summarizing thinking tokens since January to counter distillation.
Do I really have to explain the shape of AI training pipelines to you?
Train a big, wide base model with a lot of potential. Mid-train or post-train that on Claude Opus 4.5 reasoning/agentic traces (i.e. Claude Code data from Chinese API resellers) to make your model approximate a high baseline of chatbot behavior, reasoning, agentic work and tool use.
Then run your own expensive SFT, RLHF and RLVR on top of that yoinked baseline to dial it in further.
Actually doing RLHF and RLVR is extremely expensive. Distillation gives you a lot of dense, high quality post-training signal for cheap. This can get your model into the basin of "the right way to tackle this kind of problem" without a frontier lab compute budget. It's a big shortcut that gets you closer to the target - you can take it from there and build on top of it with your own work.
Also, it's unclear whether "summarizing thinking tokens" actually ruins distillation, or just makes it work worse. I'd bet on the latter, really. Because it's an approximation game, and summarized reasoning is still a better approximation of true reasoning than most of what you get online and in pre-training datasets.
3.4 million is the number of sessions Anthropic detected. The actual number of Claude sessions trained on is likely >100 million. There are tens of thousands of accounts funneling Claude sessions into Chinese labs https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
They are used for post-training, i.e. calibrating the model to understand and use tools/command line more effectively.
> 3.4 million is the number of sessions Anthropic detected. The actual number of Claude sessions trained on is likely >100 million.
That's an increase of only a single order of magnitude, increasing my estimate of exfiltrated tokens from 0.05 to 0.15 trillion - a far cry from the 15 trillion required.
> They are used for post-training
Possibly - it may be too much data for post-training, unless further curation was done. However, this is not distillation; you know it, I know it, Dario knows it, but "Distillation Attack" is a short, memorable, sciencey-sounding, political sound-bite with enough malevolence to be deployed on the floors of congress, or by the usual fear-mongering newstainment talking heads.
You're conflating pre-training data volume with post-training data volume.
Nobody is suggesting Moonshot used 15 trillion tokens of Claude data to pre-train a base model from scratch. That would be impossible and nonsensical.
This is entirely about distillation, which happens during post-training (alignment and SFT). Here, datasets are measured in millions or billions of tokens, not trillions. 50 billion Claude tokens is far, far than enough to copy Claude's reasoning logic, writing style, and tool-use ability to the pre-trained base model.
> However, this is not distillation
I don't understand how you're so caught up on the term "distillation". Distillation is using a larger model's outputs to train a (weaker) student model. Which is exactly what's happening. It's a standardized term that has been in use for a decade.
There is a lot of supposition going on your part and mine. IMO, Chinese labs are not dependent on OpenAI/Anthropic outputs; they definitely use the outputs, but along other training/post-training data.
Now that Anthropic hides the real thinking tokens in a way that precludes future CoT distillation, we'll find out which side is correct based on whether Chinese AI labs close the gap or not.
My bet is they'll close the gap; nothing about frontier AI is magic, once something is shown to be possible, experienced practitioners almost always figure out how to accomplish the same feat, though not always on the same way. This is why frontier US labs keep leapfrogging each other every few months.
or the application layer - which will capture majority of the value.
yeah hardware companies make for nice stories or green numbers on Wall Street - but value will be captured by application layer.
look at history.
That’s true up until the point where you can ask the hardware you made to make its own application layer.
assume you are a "second class lab" and you are in fact making progress by distilling the results of the frontier labs' efforts.
what is the end game for this strategy?
if the frontier labs shut down, or stop releasing to the public, and there's noting left to distill, how will you progress?
This line of thinking makes no sense because it assumes that labs that distill from frontier models are doing nothing else. It's the classic "the Chinese can only copy" mentality, and it's going to end poorly for American companies.
I'm pretty sure that all labs are distilling each others' LLMs, maybe apart from Anthropic and OpenAI. It would be stupid not to do it, because it's cheap and effective. But that's not the only thing they're doing. If you think K3 and GLM-5.2 got this good only from distilling frontier models, you're not paying attention to Chinese labs' publications.
i never assumed that, and i do keep up with the publications. i'm also not saying it's a dumb thing to do! what i am saying is that empirically, it appears that distillation of a more advanced model is a required first step for them to train a borderline competitive, cheaper model. in effect, their training is subsidized by the frontier labs.
if this were not the case, then we would be observing chinese models that far surpass frontier models in capabilities, rather than "almost as good, but much cheaper", and we would be having a very different conversation. what happens to these efforts when the subsidy is cut off?
> empirically, it appears that distillation of a more advanced model is a required first step
I see no evidence for that.
> if this were not the case, then we would be observing chinese models that far surpass frontier models
It's pretty clear that the primary reason for the difference is budget and compute availability. Chinese labs have at least an order of magnitude less money than Anthropic and OpenAI.
> what happens to these efforts when the subsidy is cut off?
They will continue making progress as they do now, minus the benefits of distillation.
https://www.anthropic.com/news/detecting-and-preventing-dist...
Moonshot AI Scale: Over 3.4 million exchanges
The operation targeted:
Agentic reasoning and tool use Coding and data analysis Computer-use agent development Computer vision Moonshot (Kimi models) employed hundreds of fraudulent accounts spanning multiple access pathways. Varied account types made the campaign harder to detect as a coordinated operation. We attributed the campaign through request metadata, which matched the public profiles of senior Moonshot staff. In a later phase, Moonshot used a more targeted approach, attempting to extract and reconstruct Claude’s reasoning traces.
I'm assuming you posted that as evidence for the claim that "empirically, it appears that distillation of a more advanced model is a required first step", but I don't think it is. It's just evidence that Moonshot distills Anthropic's models, which, yes, they do.
it is not a required first step for training a model, sure. but that's not what i claimed. what i claimed is that is how they are so significantly _reducing the cost_ of training one! how else do you think they are doing it?
>request metadata, which matched the public profiles of senior Moonshot staff
Translation: we have the machinery in place to identify our users, and actively do so.
Distillation from a teacher model solves the self-start problem, that is, building a model to the point where it reason coherently. Without distillation, solving self-start is incredibly difficult since it requires millions of high quality training samples. Creating that kind of dataset takes an enormous amount of effort.
Once a model becomes competent enough to perform complex reasoning, a teacher model is no longer necessary. The model can now reason about its own behavior and build a better version of itself through recursive self-improvement (RSI).
Kimi K3 is capable of RSI.
In public with budgets that don't risk destroying the American economy presumably. Yes it may be slower.
> with budgets
and what will fund these budgets exactly? inference is cheap, distillation is cheap, training is what's expensive.
Same people who fund linux kernel development. A coalition of companies that find it useful.
Presumably the US military / NSA.
the USG/NSA will fund chinese labs? to what end?
I was more thinking they would be funding US labs.
the question was: what is the endgame for the stated "second class labs" strategy of distilling their frontier competitors then undercutting them on price?
They will make a bunch of money then maybe go out of business eventually when the economics shift? Do they need an endgame?
Some people are just happy to follow the money.
Yes yes, we all understand the game-theoretic race-to-the-bottom you're describing here. Somehow despite linux being FOSS it still powers most of the important computing in the world. Can you explain how that works despite it being free? Once you understand that case I think you'll understand the game-theory behind how large projects can exist in the absence of traditional IP protection.
the obvious difference is the massive scale of data and compute required to develop and evolve these models, and the costs they impose on those building them.
Smaller budgets, slower improvement, less risk. They're not entitled to profits if that business model isn't sustainable. They're not entitled to a change in IP laws to protect their business model. They're not entitled to growing that fast.
who are you talking about? again, my question is concerned with the "second class labs" and the sustainability of the distillation-as-a-service model.
Making lots of money?
There doesn't need to be progress at this point. Some models even from 1 or more years ago are useful for some purposes
Well, there is precedence: Google can scrape the web, but you can't scrape Google. Laws around compiled databases exist for a reason: you can't just copy the phone book if effort has gone into compiling it, it is itself copyrightable
That varies by jurisdiction. In the United States, copying the phone book (or otherwise copying facts from someone else's collection) has been legal since 1991:
https://en.wikipedia.org/wiki/Feist_Publications,_Inc._v._Ru....
And funnily enough, most laws about compiled databases might not apply, for one.
And then there's new updates related to AI that fully take out LLMs from protection.
This is the opposite of legal reality, at least as far as the US is concerned.
> Distillation “attacks” are not attacks.
Say it louder for the people in the back. All these complaints about "distillation" from frontier labs are bordering on felony contempt of business model at this point. It's great for us. Maybe it's bad for them but nobody other than shareholders really cares.
The optimal outcome for humanity is for oligarchs to spend trillions training a godlike AI, only for the precious weights to just leak. No "distillation" required.
The hand wringing over whether internationally located AI labs are "stealing" output from American ones is the funniest thing in a while.
It's international politics with people talking about AI success as a matter of national strategic advantage and survival. So at best "this was built off our work" mostly tells you that apparently you've got months of advantage when a new model drops before it can be cloned. That's certainly some sort of advantage, sure hope it represents a consistent ability to stay ahead and causes people to redouble their efforts.
Or...of course none of these companies are worth what they say, but the advantage is also not really that great, and a whole lot of people are just really worried about their stock payouts.
[dead]
Thanks for the models guys, sorry for your losses. Once this reality becomes mainstream and undeniable, surely the bubble pops and then what then. Future model development stops? Becomes private? Becomes a public effort?
The existing models are still going to exist. As hardware improves, there will be a day where it might cost a tenth of a penny to churn through 100M tokens a second of Opus 4.8. Established compute providers will invest in improving the models incrementally when margins drive them to look there.
> Distillation “attacks” are not attacks.
If "distillation attacks" happen, we have to conclude there is some value add in what model labs do. Regardless of how we feel about using existing human knowledge in the way they currently do, it's simply impractical to infer that everything that happens downstream of LLMs can not be an attack on some IP because of it.
So both things can be true: a) People infringe on Anthropics IP and b) what Anthropic did to build their models is legally questionable (or might be ruled illegal, even though I doubt it).
>People infringe on Anthropics IP
No.
Authors do not infringe on IP when they read another's book, nor should the lumber company be able to dictate how I use planks and if I can resell them if i'm done with them.
You're framing it as if the added value of the author or lumber company, awards them consideration when somebody uses the products to create more value.
IP law was always a big mess, and these questions cross far into ideology instead of law; but I do not understand people who think we need an ideology where more IP-law is good for society.
It's more simple: They infringe on the IP by way of violating the ToS. If you violate ToS and the company suffers financial harm, they usually can (usually) sue you in civil court for damages.
Violating terms of service and violating IP rights are two independent violations. Neither implies the other.
You can't violate ToS you never agreed to. If I use pirate Claude through a third-party reseller, I have entered no agreement with Anthropic.
Api key?
I guess you could steal them but thats a whole other issue.
Terms of service are separate from intellectual property
That’s not what “IP” means. You’re describing breach of contract.
>Authors do not infringe on IP when they read another's book
Are the distillers reading books or are they building models?
If anthropic is providing no value they can just build from scratch. But obviously distilling is easier. Hes saying thats the value they add.
There are some quite interesting legal implications here. If Anthropic has IP over output produced by agents, do they somehow have legal rights to code and documents produced by such agents?
This would demolish agent usage by corporations.
You say “if”. How did Anthropic obtain this IP, if the model serves ripped internet and all human knowledge?
General consensus is that neither the model nor its outputs can be protected IP
> People infringe on Anthropics IP
Unless someone literally stole the weights somehow (which is not out of the question, I doubt either oAI/Anthropic have the capabilities to prevent a state-level actor getting those weights), distillation from generations is not infringement on anyone's IP nor is it stealing nor is it an attack. It can't be. As long as you pay for tokens you get to do whatever you want with them. Someone saying you can't doesn't mean it's an attack or their IP or whatever. They either sell the tokens or not. They can decide to not sell them to anyone, but again that's not stealing.
And their ToS are a joke. Imagine how people would react if MS had ToS saying that you can't use MS software to develop solutions that compete with MS. They'd be laughed out of the room. Somehow it's ok for token sellers to decide what you do with the tokens? Why? If you pay for something you get to do whatever you want with that output. Train, distill, whatever.
Its definitely an attack. Thats established from anthropics perspective. No one has a right to use Anthropic’s services in ways that directly violate the ToS and user agreements.
There's a lot of people diligently shilling for Anthropic in this thread. That's established from my perspective.
> from anthropics perspective
I guess I can see that, if you mean the targeted effort of creating many accounts w/ the intent of doing it at scale. Sure, they may see that as an attack. But again, it's only an attack from their perspective if you agree that using generations to distill is "wrong". I just don't see it, in general. You can't both sell tokens and decide that distilling is somehow illegal. Something, something, cake and eat it.
> Its definitely an attack. Thats established from anthropics perspective.
How do things get "established" from someone's perspective, exactly?
By that logic it is established from my perspective that Anthropic has no right to train on anything I've written that is publicly available on the internet.
Of course, they don't care about my perspective, but then again I don't care about theirs.
> People infringe on Anthropics IP
Anthropic’s model outputs contain no IP. This is actually a simple legal proposition (rare in this field!) that derives from the fact that only specific classes of IP exist: copyrights, patents, trade secrets, and trademarks. Examining each, it is clear that API outputs do not qualify. Anthropic disclaims copyright in outputs; the outputs are not patented; the outputs are not secret (a prerequisite to having trade secrets); and trademarks are irrelevant in concept.
The output of Anthropic's models is not Anthropic's IP, as that would destroy their market, if Anthropic owned all the software it generated, and all the content. So distillation, which is just using those outputs is always going to exist.
I'm pretty sure that LLM output is not intellectual property. Nobody owns it, and it can't even be copyrighted. So using output from Anthropic's LLMs in ways Anthropic does not condone is not IP infringement.
Whether or not its legal to distil models, it is obviously morally permissible to do so.
Anthropic, OpenAI, etc do not deserve legal protection.
anthropic model output is not their IP
that would be existential doom for them because then they have a case to claim ownership of their users' codebases
no corporation would sign off on that
The value is simply that it is easier. The same way it is easier to ask someone who has experience for advice than reading hundreds of textbooks.
Considering they were the original infringers, I don't know how anyone can expect tears to be shed here. The best we can hope for is for all these cancerous - and they really are the definition of a cancer - money burning entities to all fall apart to distillation attacks like these.
Regardless of whether it’s intellectual property or it isn’t intellectual property, it doesn’t actually matter. If AI doesn’t stop seeing diminishing returns in scaling up, and it hasn’t yet in the 10 years since the attention/transformers paper, the advent of AI will be the most important development in the history of humanity. Controlling that machine, or at least having one of your own, is an existential problem for nation states. It’s like a matter of national defense.
Do you really think intellectual property laws will prevent this in practice? It’s like as if we said, “hey, USSR, you can’t make a nuke, too! We patented that already.”
Asking China to not distill our models down is equally as ridiculous.
I do like that you mention diminishing returns, because we are hitting them in building out all the external requirements for competing at the frontier. Even if model performance scales linearly with energy input, the top labs are now competing with other uses for that energy.
How far are we willing to go as a nation (and as a species) to prove out the scaling laws? Are we willing to sacrifice our industrial base? Would we rather train models or smelt aluminum?
It's unlikely the USA would be granted an exclusive patent for the atomic bomb given the well-established existence of prior-art in the form of nuclear fission on the sun.
(I actually appreciated your analogy, despite my lark)
In fact, China stealing fire from the gods is essential to the future balance of power, so long as they keep making the results freely available.
Anthropic’s IP is basically null and void for how they created it. And they might not want to try and challenge this in court, considering how they had to settle for using text books they had no right to use
>Distillation “attacks” are not attacks. The frontier labs “distilled” all existing human written knowledge into their models
So why didnt we have these LLMs in 2005?
Answer the question "how much does 5 cents of LLM computation in July 2026 cost in July 2005" and you'll have the answer to your question.
Don't forget to account for all the costs. It's not just that CPUs are X times slower. Memory is X times smaller, too, and networks are X time slower. And all this hardware is many times more expensive.
If I'm getting my mental estimation right, training a 2026-frontier-class LLM in 2005 would be somewhere on the order of all the computation power in the world at the time. It's not that many more factors of magnitude before you end up at "all the computation power in the world up to that point".
Is this some form of rage bait? 2005 we hadn't the GPUs, we have today. There are other factors, but I think this is the big one. The mathematics of building an LLM are really old, we just hadn't the hardware to do the needed calculations.
Right. Therefor it's not simply a derivative of information. The hardware is required to build the model. Software as well. The model uses information, it is not "distilled" from it.
"Distillation" literally means to separate and take some components out of something. You can distill how a model works from a model. You cant distill a model from information because the information does not contain the model.
People are happy to conflate distilling with building because they dont like how the information was used. You distill how the model works from the model, and you build a model with information. Both could be morally good or bad but its not the same thing.
Information is information. Why is some information considered different than others in your estimation?
> Information is information
Not really, what the information actually is, matters a great deal. It's harder to get good results going from "nothing > model+weights" than "nothing + traces from known good sessions of other good model > model+weights", this is what the "distillation" part is referring to. If "information is information", you wouldn't even need to separate good from bad sessions while doing the training, which leads to somewhat obvious results if you don't.
Can you be more specific? I have no idea what you are trying to say.
To succinctly restate my point, you cannot distill a model from information because the model is not contained within that information. You can distill a model from another model.
Their point is that "training" and "distillation" are essentially the same. The difference between the words is whether the source material is output from another model, vs being some original text.
That argument is moot as distillation also requires a lot of hardware and software, if copying models was as easy as that, we would have hundreds of competing models.
No. Building models and distilling models both require the hardware and software. It doesn't mean building models is distillation.
Because the transformer architecture that enabled modern LLMs wasn't invented until 2017[1]?
1: That's the "T" in GPT fyi, even though Google is the author of the research paper that changed everything
Right. So we had enough information to train LLMs but not the technology to build it.
So the initial models arent just distilled from information. We’ve always had the information.
Yes, but not the secret of distillation.
Moore's Law or something. Were you alive in 2005? The Nintendo DS getting the Opera browser was a big deal. THAT 2005 with today's LLMs? Hilarious.
We didn't have the compute required (GPUs powerful enough to parallelize forward and backward pass). This compute is what allows us to train from human knowledge or distillation.
because you had neither the chips or the information in 2005. You have probably on the order of 5000x to 10000x more GPU compute today than you had in 2005 and three to four magnitudes more openly available data.
The first "L" in LLM does the work. In 2005 you had no Github, Stackoverflow, Youtube, common crawl and no archive of digital ebooks.
> There was never any plausible explanation for why this wouldn’t happen.
What a nice post hoc revision of history. Distillation is still an active area of research, that you can distill models as easily as you can it genuinely interesting and absolutely not something that was taken for granted even 12 months ago.
Even 6 months ago this idea that 'using model outputs as training examples' was listed as the reason that all models would fail in the near future due to some spooky circular training catastrophe.
Don't pretend like this was so obvious.
I think you’re being overly combative. It’s intuitively quite obvious that it’s incredibly easy to implement and the circular training catastrophe was only ever a conjecture. It’s kind of like releasing a crypto primitive without knowing a proof. Like… maybe it works, but you can’t assume that just because you don’t know how to break it. You have to remember that 100s of billions of enterprise valuation rely on frontier models being moats. The burden of proof is on those raising valuations assuming they will capture the full market.
I agree that hindsight is doing work here, but DeepSeek R1 from Jan 2025 seemed to heavily leverage distillation, and 18 months is an eternity in this climate.
This was always where this was heading, but we got here much faster than expected.
Once western governments declare it to be a "national security" risk for citizens to have access to open-weight frontier models, and once they classify using these models as acts of terrorism, what will that world be like?
Will using Kimi K3 come to be like how napster was in the olden days? Everybody knew it was technically illegal, but come on -- any track at your fingertips? But surveillance is quite more evolved now.
Or it will be like cannabis, where a guy in the neighborhood will low key rent you metered access to the 8x5090 rig in his basement he cobbled together from parts on ebay? Or everyone will flock to VPNs?
Or will the oppressors actually succeed? The same way that napster is long gone, and everyone accepts that they must pay spotify for a homogenized collection, where artists must take only a minuscule cut (more than napster though)... We'll be stuck with nerfed Cohere or Mistral models for open-weight options, as if they need more lobotomizing. Or else we can pay through the nose for Anthropic/OpenAI for "American Frontier" models which will fall increasingly far behind China.
Or else, like how Kindle Fire was subsidized by ads, we'll have "Kindle AI" where influence is sold to the highest bidder, where the LLM will tell us that smoking is actually healthy if big tobacco can engineer its renaissance by turning its lobbying dollars to pay-to-play, pumping its propaganda into the training pipeline for Amazon's extra commercialized line of ultra budget LLMs.
Basically a new iron curtain didving the world into digiatl blocks.The era of open internet/science is on its last legs with the potential forr bifurcation into incompatible ecosystems high , the onger the exchange is disrupted. As recently as this month the USgov has donce a Wolf Amendment style declaration for the Scientific collaboration NSF while shifting its purview under the military. To add to that its trying to rope as many countries into its Pax Silica idea intentionally to exclude China while simultaneosly coercing its 'allies' into using its nerfed offerings [1]
So maybe some isolated switzrland/singapore type locales would exist for US/EUusers to be able to dip their toes across the curtain legally without reprucursions.
[1] https://nitter.net/RnaudBertrand/status/2069574934972797089
> The Wolf Amendment is a law passed by the United States Congress in 2011, named after Representative Frank Wolf, that prohibits the National Aeronautics and Space Administration (NASA) from using government funds to engage in direct, bilateral cooperation with the Chinese government and China-affiliated organizations from its activities without explicit authorization from the Federal Bureau of Investigation and Congress
https://en.wikipedia.org/wiki/Wolf_Amendment
At this point, the United States will lose that battle most Countries in the world are going end up using electronics from Asia, that ship has sailed Japan, China, Korea, Singapore, Taiwan, Vietnam, dominate that area, China already dominates EVs, Drones and many other electronic devices, and with the way Donald Trump has picked fights, Europe, Canada, Australia, New Zealand, Mexico and many others are looking for other business partners.
If you need infrastructure done, China is dominating that area too. Rail, High-speed rail, Nuclear reactors, (near future Thorium reactors), Dams, Highway roads, bridges, Ocean ports, airports you name it, and they can roll it out, Transport ships, And if they don’t do it, Japan, Korea, Vietnam, and Taiwan do.
Is it too late? No, not necessarily, but America needs a regime change…
Singapore? Singapore doesn't produce shit.
Are you dissing my boy Sound Blaster??
[flagged]
We've banned this account for using HN for ideological battle and ignoring our requests to stop.
Hilarious response to concerns about industrial production, thanks for getting us here.
If rural America is that unappetizing, you understand you can just go live somewhere else, right? There is a very deep-seated hatred here that I suspect has little to do with actual "rural people".
America is what it is. The only thing that will change it is leaning in not bemoaning "rural people" on Hacker News.
If they outlaw open source models that'll just handicap American companies, the rest of the world will be running open source and have an arbitrage against US companies.
Exactly. It's an incredibly stupid idea to restrict access to open source models if your want your nations economy to succeed.
> western governments
Are you talking about the US, specifically?
Why would other countries, that don't share the same anxiety about China as the US, would be troubled with the this?
"Why would other countries, that don't share the same anxiety about China as the US, would be troubled with the this?"
It's the other way around.
There is a high likelihood that many countries of the "west" (the "global north"?) will outlaw, restrict, or otherwise control LLMs and the tools that enable them.
The US, however, is blessed with the first amendment which makes it extremely difficult to restrain speech in any form - including code.
> There is a high likelihood that many countries of the "west" (the "global north"?) will outlaw, restrict, or otherwise control LLMs and the tools that enable them.
Why? Based on what?
I've seen absolutely no indication of this.
Only the US are playing this game atm.
It wasn't difficult for the US to restrict TikTok (or BYD, Huawei, DJI...)
[flagged]
The US might pressure them?
US pressure is worth a lot less than it used to be; that's why other developed countries are urgently prioritizing digital sovereignty after years of technological sclerosis where they were happy to run on US-managed cloud infrastructure.
It's not just the tariffs and imperialist/autocratic aspirations of the current President; it's also the fecklessness of the federal legislature and the revelation via social media that a large cohort of the public hold a negative-sum worldview and enthusiastically endorse bad faith dealing.
It’s worth less, unfortunately the country is still the most powerful military and economical actor in the world, and still has a lot of influence
EU countries are homeshoring their digital stacks as fast as they possibly can, and the reason is precisely because of the pressure that the US government is exerting via its (temporary) dominance in technology.
That’s really something that has to be seen, we (European countries) talk a lot about sovereign software but have very little to show for it. Things are moving but it’s still mostly a political posturing more than anything else. Might be different in a few years
Those days are gone. Look no further than the occupant in the White House. IE the Swedish jet industry is about to get bigger, future drone expertise, if the Ukraine can hold on if you want to learn the ins and outs, you don’t need the United States. If you’re serious about learning and building drones.
It’s going to be a different world, a world where many former allies are not gonna look to the United States first they can no longer afford to.
It’s not gone yet, the US is still a bully with a lot of power and with access to quite a lot of levers to pressure other actors. Actually, our (Europe) weak response to US aggression and threats has been disappointing so far. We will see how that evolves, the EU is slow to react
Not sure if you noticed, European countries are distancing themselves from the US. They couldn't be pressured to offer logistic support to the US shitshow in Iran, why would they be pressured to help the US in its protectionism of its AI bubble?
Im aware, I’m in Europe… The US is still a bully and I would expect to be very likely to pressure other countries, even if the influence reduced. But you will notice that I used a conditional in my comment, to soften it
Well, many EU countries like Italy and Germany officially freaked out about DeepSeek, ordering it to be banned from app stores etc.
That's about the mobile apps specifically, and it is related to personal data of EU citizens being transferred to servers in China.
It has nothing to do with running open models, especially in hardware within Europe.
The disappearance of high ram Mac studio rigs is probably just a coincidence, right? :/
Apple, like everyone else in the industry, doesn't have enough DRAM. For every 512GB in a Mac Studio, they could put those chips to 64 Macbook Neos^.
Apple benefits enormously from on device AI (sells hardware) and prominently features software like LM Studio in the marketing and press releases of their new hardware.
^Technically the on-chip packaging of A-series processors make this a bit different, but point still stands.
More like the pricing was too turbulent. Apple doesn’t really do rapidly changing prices and there isn’t any stable price for that much ram.
even 8x rtx pro 6000 is only 768GB of VRAM. IDK how anyone is going to run k3
Usually offloading experts to system RAM. DDR4 has gone up a lot, but on a 8-channel used Xeon motherboard or whatever, you can get tolerable mem bandwidth out of it.
on the 1.5 tB macStudio that is going to be released next quarter
This is why God gave us 1.58-bit ternary quants?
1-trit*
Free server racks for everyone when the bubble bursts!
No they will ship it to China as ewaste and then sell it back to us.
Almost free server racks for everyone when the bubble bursts!
I just want an Oxide server at home for sub $1000. Or this, whichever's cheaper.
"Free server racks for everyone when the bubble bursts!"
I actually have this trophy from the previous bursting bubble ... a Sun microsystems rack populated with three e4500.
$750k + of equipment at original list price ...
> Sun microsystems rack populated with three e4500
That's super cool; I bet that's a lot of fun to play around with. I wonder how much of this stuff just ended up in a landfill because it was too much effort to find buyers.
> Or it will be like cannabis, where a guy in the neighborhood will low key rent you metered access to the 8x5090 rig in his basement he cobbled together from parts on ebay?
https://www.youtube.com/shorts/iNotXHO8NWU
https://www.kickstarter.com/projects/groove-thing/the-worlds...
this doesnt even capture it.
cars are a final product. beyond a basic threshold they dont influence other products, their quality or cost structure.
this is different.
all software, and everything that depends on software, will be enshittified into trabis and ladas once the idiot west starts "building the wall".
as a developer, techie or just a modern person, the only response to this is to get out while you still can, like east germans in 1950s.
I tried Kimi K3 on a task I've done with every other model I use regularly (https://swelljoe.com/post/i-let-every-agent-implement-its-ow...) and found it chewed a lot longer on the problem and ate up almost the entirety of a 5 hour usage limit on their $19 plan.
I only have the $20 plan from OpenAI and the same task, with a lot of the same implementation details as Kimi Code, only took a few minutes and consumed almost none of the 5 hour limit.
Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare, but when I sat down to add Kimi Code to flar, it was because I wanted to try it on some real work and then couldn't do any, because usage was nearly gone after the trivial task...no other ~$20 subscription I have has felt that tight before.
So, it was really slow to complete the task and seemingly much more expensive than every other model I'd tried. Maybe bad luck. Maybe it'll do better on other tasks. I wouldn't know as I was out of usage when I had time to try.
It did find a bug that Gemini 3.5 Flash introduced unprompted, though, so it has that going for it.
We really need to stop using $/M tokens as the pricing benchmark. I've found that the number of tokens used tends to be a bigger factor than the listed per token price. The cost per task vs. intelligence curve is really what you care about, and in my estimation Chinese models are just not there. They are focused on benchmaxing and getting the highest raw score they can, rather than efficiency.
The artificialanalysis cost per task chart has DeepSeek as the clear winner and Fable as the clear loser. But I would still pick Fable for some tasks, so that also can't be all there is to it.
But I agree that price per token figure is not great. It seems even the tokens per character can vary between models, so it's basically useless.
Wow, you weren't kidding. I looked at their chart, and the cost-per-task for Fable is more than double Sol's. And DeepSeek absolutely stomps. Four cents per-task vs Sol's $1 and Fable's $3.
I might need to check out DeepSeek more. I had no idea the difference was this obscene. Makes me wonder if something's off with the benchmark. A 70x cost reduction vs. Fable seems too good to be true.
Almost as if centuries of passing the imperial exam by either skill or cheating influenced Chinese culture a lot.
While I do agree that cost per task is what customers should care about, and not cost per token. Cost per token is an objective metric. Cost to do a task can vary a lot. Different tasks, different prompts, or just pure randomness nature of models make it a bit harder to define this as an objective metric.
Unless the cost per token is prohobitedly high, people can often try the model out themselves and make a subjective judgement of how effective and efficient is it at solving tasks they usually deal with, using their setup.
Ockbench: https://ockbench.github.io/ show token usage and accuracy, but you have to compare it with other stats (prices and speed).
Yes, this is already accounted for in many benchmarks, but without deep context of the problem type, the top line pricing is the best starting point.
In my own experience, Fable is more token efficient than opus 4.8 with a higher likelihood of completing tasks correctly or at least with minimal corrective work. Opus regularly struggled to gather the correct context and reason effectively about what it had gathered.
GPT-5.6-sol crushes fable in speed and token efficiency and is clearly superior across many tasks that matter for me.
I also find all models from anthropic after opus 4.6 to suffer from the same ai slop language that long plagued OpenAI and seems to have been reduced drastically in 5.6
Don't forget that you are not really seeing the thinking tokens used - so non-trivial to count them.
In my experience, Kimi just tends to think a lot, with the main thing that takes up a lot of space is it constantly second-guessing itself. I've watched it do paragraph after paragraph of "Wait, actually..." while it stumbled and used a ton of tokens on one small detail of what it was asked to do. Though I also gave GLM 5.2 a task to port some JS code to Python to test it, and in my experience it doesn't second guess as bad as Kimi does, but it really did there. It kept doing web searches and second guessing tons of tiny little things, using $0.25 of API spend in total to port about ~50 lines of JavaScript. It did produce an error the first run, but on second run it gave me a program that ran.
I gave Claude Code/Fable the same task and it took significantly less time, but also stumbled on the same error as GLM. I didn't have it fix it though. I was mostly interested in timing differences.
I do like open models where I can, but I'm really hoping they get trained to second guess less. Or maybe I just need to prompt them differently. I'm not sure.
Interesting. Could you not tell it to not second-guess itself (on minor issues)?
Fighting against the weights is often a lost cause. For any model.
Yeah I've noted this behavior with best in class open weight models. They said K3 would have token efficiency improvements and I was hoping especially solving the thinking loop issue that plagued K2.x but even if this release helped somewhat, it looks like we still have a long way to go here... I'm not sure what's up here but I suppose lacking finetuning quality.
What OpenAI in particular have done with reasoning efficiency in the past few months since ChatGPT 5.5 is nothing short of remarkable. It's overshadowed a bit by the benchmark game and the Fable hoopla.
Now is the time to focus less on token cost and intelligence, but tokens to solve a particular set of tasks in closed benchmarks for a variety of categories.
What is the use of grand intelligence if it either costs you a kidney or can't complete at all within a token budget? Even if there are niche uses where you truly want "maximum power" above all, we need to at least more severely penalize such models versus those that does it just as fine within a tenth of the token cost.
I'm aware of some benchmarks at the Artificial Intelligence site, but CLEARLY we are not focusing enough on these today and still leaving the fun surprises to the users.
Yeah, I'm finding I end up switching to Codex and GPT 5.6 a lot lately because I've either run out of Fable usage or Fable refused to do the task. Most recently it refused to work on a WiFi configuration UI for a robot. No idea why it thought that was related to security, biology, or some other sensitive topic. They've hobbled it with guardrails that are overzealous and now there's a big opening in the market. Fable may be the best, but if it won't do the job half the time, it stops being my go to model as I don't want to waste time only to find it refuses halfway through.
Earlier today I made Claude code implement a feature with fable. It worked roughly 60 minutes and used around 30% of my 100€ subs 5h sessions.
Then I typed /code-review in a second terminal/clean session after the analysis was done (no code changes) the usage was 99%. I then asked it to write that into a review.md so I could restart from that the next day. Sadly the last % wasn't enough for that.
Ymmv, these models behave very differently with no discernable reason. Usually reviews(even with fable) take like 10-20%... Yet suddenly you get it to burn through 65-69% in 15 minutes or so
/code-review in Claude Code spawns a lot of sub-agents (counted like 8 once), each looking at the code from some certain aspect (like correctness, maintainability, duplication, testing, etc). It eats tokens like crazy doing that, but also covers quite a lot. The default code review in Codex does far less (feels like it's only correctness) and doesn't uses subagents. Actually I made a skill for Codex that does a review closer to what Claude does by default, but using like 4-5 agents and some being cheaper models/less than xhigh reasoning. I'm getting pretty nice reviews with that that cover more than just correctness.
Read the session file of the reviewer tomorrow. Sonnet should suffice.
> Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare
AI subscription pricing is so goofy. You get some amount of usage that varies by models, is measured by opaque token usage, driven by how many tokens the (usually) vendor-provided interface (or model itself) wants to use. Then your usage is limited by time opaque time windows.
You call it goofy, in a different context we would call that a dark pattern, shady, prone to fraud
AI subscription pricing was fine when it was $100/month for some opaque 5 hour token budget I don't think I ever used, not even that one day where I coded for 14 hours non-stop using Fable. But like most people with low token usage, I had a human in the loop and and I didn't use workflows with swarms of agents.
Now, of course, the plan is to remove Fable from the subscription. To paraphrase Darth Vader, they have altered the deal. Pray they do not alter it further.
Gpt 5.6 is still like this at least for the $200/month option. It’s also always faster than fabel. Fabel might be able to do some things better but I don’t have time to constantly wait and find out.
They're probably not going to remove Fable. They just extended it for another month.
Kimi K3 only supports "max" reasoning effort right now, but they plan to enable other levels soon [1].
When I looked at traces from benchmarking, I saw a lot of backtracking and uncertainty while reasoning ("wait, but..."). This also happens with GPT 5.6 and Fable with xhigh/max thinking, albeit to a lesser degree.
I think that explains part of the token inefficiency. Hopefully it will improve with lower reasoning effort settings.
[1]: https://platform.kimi.ai/docs/guide/use-thinking-effort
Absolutely do not pay for the kimi plans thinking they will be cheaper. If you sign up with a Chinese phone number, you can get the same plan for 200 yuan instead of 200 usd, it also only accepts Chinese payment methods iirc. So the plans are really made for Chinese userbase.
That sounds complicated. I'll just use my month of Kimi and then cancel. I have too many AI subscriptions to use them all, anyway. I subscribed mostly to test it. I mean, if it turned out to be competitive, I would keep it, but if it doesn't turn out to really excel and anything and also take longer than Claude or OpenAI models, I'll stick with them.
It is complicated, but paying for the cheaper usd plans really don't get you much usage.
Wow! Does it accept a foreign alipay/wechat pay account?
YMMV, but I'm a US citizen with a KYC'd Alipay account and I was able to subscribe for the Chinese price...
... but I borrowed a friend's +86 phone number, which you'll need to even see that price. or maybe a 回国 VPN will work.
Those exist?
I've had an Alipay account since 2006 and never been to China.
Of course they exist, Alipay is from Alibaba, think about who typically buys from OEMs/suppliers there...
No idea lol, didn't even know those exist..
how do we get chinese phone numbers?
esim
Nah that won't work. I don't know tbh, I just used someone else's number.
not gonna happen. China has strict real name verification for SIMs.
Is Kimi K3 subsidized as hard as the other models out there?
From the end of the month it will be served profitably by other providers around the world, like Kimi K2.6
Sure, but at those API rates it doesn't beat the subscription plans for OpenAI.
Not sure how the economics work for the Chinese models, but DeepSeek did the same task for a dime.
In my opinion, for the vast majority of use cases, DeepSeek is still the most cost-effective model by a mile. $10 feels like it lasts forever.
Yep, with Reasonix, DeepSeek is free real estate. Seems to just go and go for pennies.
And, DeepSeek is what I use for any task that works best with an API. It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry, and it's good enough to where I rarely have to follow up with a more expensive model or manually fix things. It's been alleged they're releasing an update to DeepSeek V4 Pro soon that improves it, which likely makes it a good fit for even more kinds of problems. It remains my favorite of the Chinese models, it's so cheap and cheerful. And, is also less aggressively censored than some of them.
> It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry
I use DeepSeek v4 Flash & MiMo v2.5 Pro. Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads. Although, I've begun experimenting with Hy3 (as an in-between Flash & Pro) & GLM 5.2 (for long-horizon tasks).
Maybe I’m not pushing them hard enough but I use Claude opus at work and deepseek v4 flash at home and they both seem about as capable. While deepseek is borderline free.
Does it matter? As an end user I really only care about 1) how much I can do in a week, and 2) how long each task takes.
Subsidies would affect 1, but not 2. But if some VC wants to subsidize my Claude or Codex or whatever, awesome.
The more important question than subsidy is what is the tokenomics of running the model. If it's inefficient to run on an nvl72 cluster (or whatever the heck has enough vram to run a 3T parameter model), and k3 isn't very token efficient, then it might not be that compelling of an open weights model.
3T at nxfp4 (which is most of it) is only 1.5TB of vram - so 8x288GB B300 or MI355 will do it if you are careful with context - maybe dp-attn? Certainly not TP. 2 of those together can easily serve it. The new AMD MI400 are at 400GB+ each, so 8x of them will nicely fit with KV to spare.
It doesn't matter if you can switch easily. It might matter if there are barriers to switching.
Subsidization could affect both of those. If you have $200B in the bank you can afford to throw massive compute at every single request; if you are less well funded, you might throttle more aggressively.
Additionally that same VC could be (read: is always) spent on developing the harness, and other infrastructure around the model, not just the model itself.
So it's apples-to-oranges when comparing a relatively new model to established competitors (i.e. OpenAI @ $900B funding vs Moonshot/Kimi's $30B FYI) because every new model they release is judged on "performance" which is not strictly speaking derived solely from the model.
It's possible Moonshot could get similar performance over time as the build out the rest of the infrastructure. We have no way of knowing how much of OpenAI/Anthropic's success is due to the model vs intelligent tooling built on top of it.
Do models know when they're being benchmarked? I also used K3 briefly and noticed it spent a very long time thinking and obviously a lot of tokens.
However I've seen some benchmarks say it uses fewer than fable which hasn't been my experience.
It would be really interesting to redo the public benchmarks for kimi k3 but token normalize the costs. Ok so maybe k3 beats fable on terminal bench, but how many tokens did it use?
Aren't they still locking reasoning to "max" pending adjustments to support shorter reasoning levels.
Found it!
> At launch, Kimi K3 will use max thinking effort by default, with low- and high-effort modes to be introduced in subsequent updates https://www.kimi.com/blog/kimi-k3
It let's me choose different thinking levels in Kimi Code. Not sure if it actually works, yet, but it says "Thinking set to high." when I change it from max.
Why would they do that? Sounds terrible
With the obligatory disclaimer that I’m impressed with what open weight models can do, I have the same experience with all of them.
The benchmarks come out and say they’re as good as Opus from N months ago, then I use it for a complex task and it doesn’t work as well as Opus from N months ago did when on similar problems.
There’s a real wow factor when you get an open weights model to do amazing things, but in my experience the gap to the frontier models has always been bigger than the benchmarks would lead me to believe.
There can be a lot of value in having the cheaper open weight models for chewing through lower complexity tasks (non-programming in my primary use case) at a cheaper rate than OpenAI or other frontier API costs. Even with those I can measure bigger gaps to the frontier models than the benchmarks suggest.
If the benchmarks aren’t being directly gamed, there’s at least some selection happening where training data or model structures are being picked in ways to maximize public benchmark performance. All of the labs know there’s immense value in having good benchmarks to show for your model because most LLM consumers are picking based on lab provided benchmark charts, not running their own evals. Running your own evals is hard and expensive.
> tried Kimi K3 on a task I've done with every other model I use regularly and found it chewed a lot longer on the problem and ate up almost the entirety of a 5 hour usage limit on their $19 plan
ArtificialAnalysis puts Kimi K3 just below DeepSeek v4 & GLM 5.2 in token use per task, which is about 2x to 3x more tokens than Grok 4.5: https://x.com/ArtificialAnlys/status/2077832879187620192 / https://archive.vn/zBbFi 2 other open weights MiMo v2.5 & MiniMax M3 are comparatively thrifty.
> Subscription usage limits are hard to measure as none of the providers tell you directly what it means in terms of tokens or anything else you can easily compare
I always put my coding subscriptions (that allow it) through "AI gateways" (Cloudflare & OpenRouter are free) which help track token use.
In my experience, Kimi & Qwen Cloud have opaque & restrictive limits, their "credits" drain faster. I now make it a point of subscribing (directly [0]) with providers that are transparent like MiniMax, DeepSeek, Xiaomi, & Z.ai.
[0] OpenCode Go, Cline, and AtlasCloud have generous limits for open weights, otherwise.
OpenAI measures token efficiency. Look at the API cost charts in their announcement: https://openai.com/index/gpt-5-6/
Even in this very thread the feedback on Kimi's actual efficacy is debated. I personally feel its worse than both Fable and 5.6 Sol, but I feel like the conversation isn't really about whether its good or not, but a backlash against the U.S governments foray into regulation. So I think people _want_ it to be superior out of anger/frustration with the current situation.
It's really good. I'd put it between Sol and Fable. I'm not super impressed by Sol's UI design skills, something K3 is strong at. Fable is still overall the fastest, most consistently well-performing model, though.
This does depend heavily on the kind of work you do and how you use these models, but the idea that K3 isn't right up there with US SOTA models doesn't match my experience.
This seems like a replay of what happened with DeepSeek. They put out v3, or whichever one it was, and everyone said it was over for US companies... then everything continued on.
The markets can be irrational for a while, but if the Chinese models are about 90% of the performance of OpenAI and Anthropic models, and the Chinese companies are < 10% of OpenAI and Anthropic's proposed IPO valuations, something has to give eventually.
This isn't just the AI race, but the end to perceived American exceptionalism (where USA wins by default). It's going to take a while for people to recognize that. Before that the markets will still go crazy, but that's not evidence things will continue on as "normal".
Continued how? I have switched most of my personal LLM coding to DeepSeek V4 Flash since it was launched.
And now 100% to a mix of K3 / DeepSeek V4 / MiMo 2.5.
It's nice not being called a terrorist just because I told it to reverse engineer something.
At work they are still hemorrhaging money to Western providers due to enterprise contracts but I foresee they won't renew for much longer. Specially of the upcoming final version of DeepSeek V4 proves to be Opus+ level.
Anecdotes. Where's the data?
https://openrouter.ai/rankings#top-models
Chinese models dwarf USA models usage. And now there's a Fable/5.6 alternative. The gap widens.
Now go and ask for your GP poster for their data as well. Unless you're only interested in data that supports your bias ofc.
I can't count the number of times I've heard people here say the frontier models are 6 months or more ahead of the open-weights models. That's not true anymore. So the goalposts are shifting.
That would make sense, what the US government has done this year with regards to AI is unacceptable
Agree completely.
When you net out across benchmarks and firsthand reviews it seems like it's maybe a little behind. There seems to be a consensus it's token hungry and a little slower. So maybe it's a point release behind.
That's weeks maybe months behind, not months maybe a year behind. It's "would my life really change if Claude was gone, not really" behind.
I actually haven't used it much, because Claude started kicking ass again the last few days. Like, way too much of a difference to be normal load-based variance. I got more done in the last 48 hours than week before that.
So, fuck yeah competition.
Just for the sake of argument and using some admittedly insane numbers, give me Opus 4.5 at a tenth the cost and running ten times as fast and I'd take that for almost any coding task over any current frontier model. There was a real phase transition somewhere in that range and improvements since then, while impressive and useful and by the benchmarks quite large, have in practice not been anywhere near as big a phase change. Honestly until we get to the point where the models don't need any checking at all, incremental improvements on how much checking they need don't do all that much for me. In practice "they get 90%" doesn't differ much from "they get 94%".
I think my main problem with the current 'aligned' version of models is that they're aligned with the very worst of the California social justice warriors.
I think it's the opposite.
Kimi K3 has 2.8 trillion parameters. We don't know the number of parameters of ChatGPT 5.6 or Opus 4.8, but it's probably in the same region. Fable/Mythos are rumored to be around 10 trillion.
So, K3 is directly comparable with ChatGPT 5.6 and Opus 4.8, and the price is not so much lower:
K3: $3/$15 per 1 Mtok input/output ChatGPT 5.6 Sol: $5/$30 Opus 4.8: $5/$25
This is not a watershed moment. It's a competitor converging to the same capability and trying to undercut your prices, but not by a lot.
As for the open weights? For now, Kimi K3's weights are closed, and I don't expect the situation would change.
> As for the open weights? For now, Kimi K3's weights are closed, and I don't expect the situation would change.
It'll change on July 27 (based on https://www.kimi.com/blog/kimi-k3):
> The full model weights will be released by July 27, 2026
July 27th. But I agree with you that this is just normal competition. The only threat this poses is to Anthropic. OpenAI is more than capable enough to out-compete, their pricing is already reasonable. Greedy Anthropic will do their very best to try and stop this though, because they want to maintain the status quo of ripping everyone off.
And how much token and time you use to solve a problem? The price alone doesn't mean anything.
Example DeepSeek-V4-Pro (high) needs 10 times more token then GPT 5.5 (medium) and can compete only with the price.
The real price saver are the cache prices, the ting, that nearly nobody has on their radar.
I’d also note that running a 2.8 trillion parameter model at scale efficiently is not simple. I would expect when open weights land getting it running fast, efficient, and at full capability will require sufficient resources it’ll be expensive outside of Chinese hosting. Which I think almost no western corporation would use for any internal work. You have to anticipate your use won’t just go towards training but will be actively mined for IP, trade secrets, MNPI, etc, or anything of use to the Chinese government or Chinese companies. I don’t say this to crap on the Chinese - but this is the playbook for the last 30 years.
That said I fully intend to use deepseek hosting for operational agents that are making decisions about non sensitive material. The economics are astounding.
Kimi? The economics aren’t that amazing to merit switching from 5.6. I expect fable will rapidly reappear in subscriptions. Competition is good.
The APIs for the frontier models via the US hosters do the exact same thing wrt saving the requests and responses for data mining. Let’s not pretend that pervasive surveillance is an eastern thing.
Given how OpenAI got rid of their 5-hour limits and reset weekly limits so often, is Kimi really undercutting them on effective price?
It was all distillation up to this point anyway. And I agree with what Suhail said on twitter: "Make the margins next to zero for all these AI models. It was trained on humanity's data, it should be gift to ourselves. Doing so will save us from a few in control of our species."
Normies still thinking this beasts of model coming from China are „dIsTIlLattiOns“ is so funny to me. Many people are not aware of the wave that‘s going to sink US „frontier“ labs that enjoyed and dreamed of stealing tons of data while making people depend on their censored and dumbed down models.
Well, there is the small issue of privacy policy: Kimi will train their models on your interactions if you use their subscriptions, and only with direct API usage (billed at API prices) they say they won't. Whether you trust that is another matter.
Those things do make a difference to some of us, even though nothing is black and white. In my case, I'll probably want to wait until other providers appear through OpenRouter and then I'll try to judge how much I trust them. But even if I don't trust them much, they don't train models anyway, so the likelihood of my data being used that way is smaller.
> Kimi will train their models on your interactions
I find these kinds of concerns increasingly silly: most of the input to these models will be ... previous output from the very same models, alongside the occasional half-assed human command to fix something and "make zero mistakes". Who cares if they train on that? Let them, if it makes their future models better!
99% of users are not working on any special IP to worry about that.
It means there's a non-trivial chance a future version of the model will know private information about you.
Maybe you're super careful with this stuff, but with agents and harnesses being given access to user data and accounts, I don't think it's feasible to actually monitor what information is uploaded and whether they involve private information.
I personally keep local models around because of this.
> In my case, I'll probably want to wait until other providers appear through OpenRouter and then I'll try to judge how much I trust them.
Keep in mind that the Moonshot team have identified multiple providers who configure their setup wrong which make their model perform worse than expected. This is why Moonshot created Kimi Verifier, but I guess its up to the provider if they want to do that.
Indeed, the B2B / no-data-retention market is still going to provide plenty of business for American companies even if every hobbyist uses open-weight models.
Exactly; this is a no-go for me, I will wait for an independent provider to sell the service, which is possible thanks to the open weights.
The right analogy here is not the auto industry, but the music industry. Regulation might "win", but margins will be driven down to commodity levels. That is not the assumption that current US AI company valuations are based on.
I wonder if this is this just the law of diminishing returns at play?
My thinking being, it's been a few months since I thought the code generation machine was the problem, rather than my interactions with the machine. A month is a long time in AI.
What I mean is, these things are about as smart as they need to be already for the average SWE. I don't think this is true for those solving the really big questions like curing cancer.
> I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart
When you say "Claude", do you mean Opus? Fable? What effort level?
This line made me think by 'normal coding work' the author means doing something they don't understand well enough to be able to distinguish the models' output.
Steve Yegge calls this is the "discernment horizon" - https://steve-yegge.medium.com/the-flat-curve-society-36c8b0...
Interesting, thanks for sharing.
Although in the month since that most recent post, his other points about open models are undercut by K3.
And at least of data available as of 2026-01, AI compute capacity was doubling every 7 months, so I expect every major country to host AI compute farms, and self-host AI feasibility to majorly increase in the next 2-3 years as well. (Partially undercutting, but not fully disproving, his points.)
And I wish his posts were 4 times less wordy.
I think a better framing is the marginal utility of the models capability growth. At a certain point frontier models will only be needed for frontier problems. The demand for that capability will decrease with time. The hand wringing about not understanding is to my mind anthropomorphic - AI of today lack agency and awareness. Even the constructed stuff Anthropic puts out there in the model docs involve contrived scenarios to elicit “scary” behaviors. It’s unclear that as models become more sophisticated whether they’re better at instruction following or not but it certainly feels that way - even if it’s through better alignment or just an artifact of scaling. However I think the malign actors of humans using powerful models for bad stuff isn’t unreasonable to be concerned about.
The marginal utility problem is a real one for AI companies. I think the current generations are already saturating marginal utility for 95% of the population. Almost everyone I know outside of my career has no use for a more powerful model. This is a serious problem for the economics of AI and semiconductor investment. This is a bigger problem than Chinese models. It leads to a demand curve problem - that supply outstrips demand.
This is comparing Fable High with K3 High. I'm mostly using these models for game development. The tasks I usually send are ambiguous visual bugs, changing the look of a scene or models, or adding a large feature. The wording wasn't accurate there. I don't use Fable or K3 most of the time. I'm usually working on smaller scoped tasks that I review myself afterwards.
Please update the blog post to clarify. “Claude” is not a model, and your writing makes no sense without specifying.
[dead]
In the future, Americans will use Chinese models and Chinese people will use American models — and neither government will be able to do anything about it.
I applaud Kimi and the open weight models - for the simple reason that they provide a price ceiling for AI.
Since Kimi’s paid plans are mentioned in the article..interested ones should know that you can only access 1M context model with $79/mo or higher plan; otherwise you are capped at 256k context. Also, with minimal $15/mo plan k3 is currently not supported at all. (prices are yearly plan discount prices)
ref: https://www.kimi.com/code/docs/en/kimi-code/models.html
Thanks for mentioning that. I also wanted to use API only and with the cache hit rates I'm getting with Reasonix/whale harness on deepseek, it's going to be a difficult adjustment moving away from practically free.
Pricing is actually far cheaper than that. There's two tiers of pricing: Chinese and US.
If you sign up with non-Chinese phone number, you're bucketed into US, you get US prices, can pay only in USD and with American credit card network.
Chinese prices are about 9x cheaper than the US prices, which are already far cheaper than Claude or other American provider. If you can somehow get hold of a Chinese phone number, keep in mind that you can save ~90% of the bill.
Most of this hand-wringing on price will go away.
My assumption is that Anthropic, OpenAI, Kimi, etc all have a similar cost structure when serving models. The same size model roughly generates the same GPU usage whether you’re American or Chinese. I’d also guess that the model sizes across all SOTA models is similar, we just only see data for open models. The difference is most likely that American companies simply charge more because they have the dominant market position.
Remember not too long ago when Anthropic was charging $75/mt for Opus? Now that many models are in “opus tier”, their pricing is $25 - higher than competitors but close. The newest Kimi is $15. 40% lower to forgo “made in America” with American enterprise support staff is not crazy. Compare AWS to Hetzner or any other flagship enterprise service to the foreign and discount option. I assume that over time, we’ll see the commodification of models reducing prices even towards the raw GPU costs.
It’s 100 yuan per million output tokens in China. That’s $14.7 USD - not “far cheaper”.
He's talking about the plans, you are talking about API prices.
The current administration's immigration policy isn't helping. This wouldn't have happened 10 years ago because the US was this city on the hill that everyone wanted to immigrate to. Talented Asian researchers would have immigrated to the US and China would be deprived of talent.
Yang Zhilin, founder of Moonshot AI, got his PhD at Carnegie Mellon and turned down US job offers to go found a startup back home in China. That was obviously a good decision, and immigration policy wouldn't have made a difference.
Your comment feels like an outdated brain drain model where talented Chinese researchers naturally want to leave China and the only question is whether the US lets them in.
That may have been closer to reality 10-20 years ago, China is a different country now, what I mean by that is they offer research funding, they have huge digital behemoths (alibaba, tencent, huawei, bytedance etc), large scale deployment opportunities and prestigious careers. Many graduates return because the opportunity set is attractive and they want to return, it's not just because US immigration policy pushed them out. Some also want to contribute to their own country's technological progress (which is a normal motivation btw), like probably you are also a patriot and want your country to succeed.
So, really, China's AI progress is not mainly the result of America failing to absorb every talented Chinese researcher. China has built a domestic ecosystem capable of producing and keeping top talent itself. I feel like a lot of Americans do not understand this.
Chinese students still want to attend US universities [1]. While it is true that the progress made by China is a factor, this administration's policies are the bigger deterrent [2] [3].
[1] https://www.latimes.com/world-nation/story/2025-02-21/why-ch...
[2] https://www.wsj.com/world/china/americas-allure-fades-in-chi...
[3] https://www.theguardian.com/world/2025/jun/06/chinese-studen...
It doesn't really address the point. Chinese students wanting to attend US universities is evidence that US universities remain attractive, not that those students would otherwise permanently immigrate to the US or that China lacks attractive careers for them afterward.
US immigration policy may be unnecessarily pushing away talent but the assumption that talented Chinese researchers would naturally remain in America unless prevented from doing so ignores the growth of Chinese universities/labs, companies, their funding, national prestige etc.
I mean, don't get me wrong, US is still highly attractive, it is just no longer the only place where an ambitious Chinese researcher can do important work and grow.
The visa that would correlate to this is the O-1 visa
20k O-1 visas were issued last FY which was mostly under the Trump admin, up from 19.5k the previous FY under the Biden admin
No it is H-1B visa. Right out of the university it is hard to recognize extraordinary talent. People like Sundar Pichai were not recognized as extraordinary right out of the university, he had to start at the bottom and rise up the ranks.
This makes even less sense, Trump admin has been here for 1 year, the implication here is a university grad on H1-B in January would become a world class researcher capable of building a frontier model in <18mo
Melania got a EB-1 "extraordinary ability" immigrant visa
To be fair, that was clearly well deserved. Marrying Trump and then becoming first lady is definitely an extraordinary ability; I doubt I could have done it.
This is why VC is actually hard. Everyone’s instinct is always “Man, once the company has demonstrated it’s awesome I would love to have been in the seed round”. The tendency to want “proven performers” is the default belief.
When people demonstrate their capability thoroughly, the Chinese government takes away their passports. You’re not exactly going to get them here with an O-1.
This admin and its policies on immigrant visas have been around for 1 year and Biden was famously pro immigration
The O-1 has also been abused for a long time, basically any software engineer kid who gets into Y Combinator has been getting an O-1
This is essentially the point of the visa, it feels wrong especially as YC drops standards and increases cohort sizes, but the same power laws that keep them winning also apply here in maximizing economic value of each O-1 approval
Basically of all visas O-1 is virtually guaranteed to have highly positive economic value
It's not the point of the visa, the O-1 is supposed to be for people of extraordinary ability, eg Nobel Prize winners. It's used for software engineers.
China never allow US AI in China, so they HAVE to build Chinese equivalents...
US immigration policy isn't a big factor.
China's got 1.8B people. If you don't think they've got the talent to pull this off, even if a lot of it leaves to live elsewhere, you're naive.
No one uses Baidu, but they built their own Google, and it's good.
They built their own Facebooks and Instagrams.
The US isn't the only place in the world where people can build software...
"China can draw on a talent pool of 1.3 billion people, but the United States can draw on a talent pool of 7 billion and recombine them in a diverse culture that enhances creativity in a way that ethnic Han nationalism cannot." --Lee Kuan Yew, former prime minister of Singapore.
hahahhahahahahaha
- thats not a sustainable strategy
- china’s homegrown tech industries already achieved escape velocity from it a long time ago, after China fenced off its market for Alibaba and Baidu in the ‘00s. some of their AI innovation at the edges was already top class 10 years ago
It has been a sustainable strategy for the tech industry for decades.
Claude is not reliable anymore with their sudden Fable access drops etc tbh and I am happy there are good alternatives coming out
I think the biggest problem with Chinese models is that they seems to overthink for most of the tasks, especially for smaller ones. The OpenAI models have in my experience only gotten better in terms of efficiency.
Yes, this (imo) is a clear result of benchmaxxing. You can get a much better score on most "intelligence" benchmarks by massively over-saturating reasoning. This looks good on those, but for actual daily usage makes the models much less effective: I don't want a model I use for coding to burn a bunch of reasoning (read: time) on trivial tasks.
It's undeniable that some of these models generate a ton of thinking tokens, but it's arguable whether that makes them "much less effective."
For example, Kimi 2.7 has been really effective for me despite having verbose thinking blocks, simply because it runs so fast. Speed-wise, it feels about like Sonnet, possibly faster.
I strongly suspect the flip side is that in the future it enables you to train smarter models by "distilling" the end result of the super duper heavily thinking models.
But these models already distill the smarter American ones ;)
It's turtles all the way down
We're having so many moments! Every day a new moment.
I have a feeling there is some elite data labeling operation going on in china for these labs at a subsidized rate somehow.
Maybe, but I don't think that's necessarily a problem. "Subsidized" is a loaded word on HN because it often refers to the unsustainable consumer pricing of U.S. AI labs that will inevitably lead to market corrections. A subsidy by the Chinese state to avoid strategic encirclement isn't necessarily unsustainable nor irrational.
Kimi K3 is really good, but it’s obviously worse than Fable, usually worse than Opus, in my experience.
Definitely not my experience. Fable is better but I'd prefer K3 to Opus based my experience with both.
That’s not obvious to me at all. Especially your claim around opus.
Agreed. I think benchmarks are pretty much right in pegging it somewhere in between Opus and Fable.
Time for OpenAI and Anthropic to close the gap by spinning up Kimi K3 on vLLM and running a distillation attack.
Terms of use are very broad and not friendly for most things.
Can’t use for commercial purposes. Can’t opt out of training. Data retained.
Correct: can't opt out of training. This is well documented.
"Can't use for commercial purposes" - incorrect AFAICT. In what sense do you mean this? The open weight MIT version obviously allows for commercial use, but I don't think that's what you're referring to, because training data is irrelevant on the open weight version. Pretty sure the API allows commercial use too. Maybe the free version doesn't? But who cares?
> Service Misuse. You acknowledge that without the written consent of us and/or the relevant rights holders, (i)you have no authority to use Kimi and the content generated by Kimi in any commercial manner; (ii)you may not use our Services to develop products or services that compete with us.
https://www.kimi.com/user/agreement/modelUse
According to OpenAI's "head of strategic futures":
1) Kimi 3 is a "very good model"
2) It's performance can NOT be explained by distillation
3) The US government should create FUD to stop US corporations from using it (so they use OpenAI instead)
https://x.com/deanwball/status/2078133895766114412
Fascinating take from OpenAI. It really gives the lie to the idea that they see AI leading to a better life for all.
"One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a 'public good' which will ultimately be provided by the state as a kind of 'digital public infrastructure.' This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end."
He never says why he thinks AI as a "public good" is dystopian, but it's not hard to imagine why. It's because he and his inner circle won't have the power to dictate what we read, see and hear.
Economists at DeepMind believe that if AI were treated as a public good—like water or electricity—with profits distributed across the entire market, ordinary people would simply be able to invest in mutual funds.
I never truly understood what the intended business model around LLMs was. Get them widespread through cheap pricing and then jacking it up? Being the only ones that had a viable product so to get the ability to extract as much value as you want from AI?
I don't understand how a product that:
- is interfaced with and is deeply linked to natural language, so everything you produce (sessions, history, etc) is in Markdown and you can literally install a second model and tell it "hey import all of Claude's memory into yours" and that's it
- is based on well understood technology, the real constraints are how much money you put into training the models, but the theory has all been developed in the open
- clearly has a threshold where it quickly commoditises and turns from "I want the best" to "hey the best is a bit too expensive. The second best is half the price and works close enough".
was ever supposed to be a money printing machine. The fact something is extremely useful doesn't imply it's extremely profitable.
IMHO we're clearly speedrunning the process of turning AI into a commodity. Dario Amodei knows pretty well that when or if Anthropic cuts people off Fable, the vast majority of them will definitely not pay for it because Opus 4.8 is good enough for almost everybody that _knows_ what they're doing, and so are basically half of the most recent models. If I already have good baking skills I don't become more productive with an automatic bread machine, I just need a better dough mixer and oven
There is no business model. That’s not a joke, the idea is to be the one that survives the race, then figure out how to be profitable. If you look at the level of capex and money raised, that’s not something you do if you have an actual business plan. We are very far from business fundamentals
> I never truly understood what the intended business model around LLMs was.
A closely related question is “what do the American labs need to do in order to justify their enormous market valuations?”
It seems like the answer cannot possibly be “gradually improve model capability while figuring out how to better monetize inference.” The valuations are just way too high for that to be sufficient.
Surely the answer has to be “continually achieve large leaps in capability comparable to the first consumer releases of ChatGPT while also maintaining a significant capability lead over open models and new competitors.”
And does anyone think that’s going to happen? Even with state-level protection from competition (which incidentally would significantly harm the American economy), the large leaps in capability seem to be coming fewer and farther between.
> I never truly understood what the intended business model around LLMs was
What appeared initially to be a huge innovation was later easily duplicated by many. There are no platform-lockins or network effects. Switching costs for users are zero, and there are low barriers to entry, with vast numbers of models to choose from and more appearing every day. As a business a token will be a commodity like an electron. Doesnt matter who produces it, or how (solar, wind, coal, nuclear etc) as long as it powers my toaster.
A lot of people still think AGI is going to happen, yes. Not the ones who actually build the thing, but the marketers above them and their eager victims in the political and business-owning classes.
It's fairly simple. Sell GPU compute + extra margins as only some GPUs can load the models + extra margins based on how much better closed source models are from open source ones + hopefully reduced cost due to batching
Same business model as always: build cool tech because it is cool and figure it out later.
It seems like the endgame is to amass absurd amounts of hardware and produce something that will replace you the baker entirely
everything else we see today is just preparing for it.
The valuation is based on one lab getting a decisive first advantage, and turning that into a durable self-improving advantage that can never be caught up to. If any can pull it off (a gigantic if), they will effectively own most AI value, and the people who own their shares will live happily ever after. Divide your investment between the labs that could plausibly do this, and your EV may not be dreadful.
This is clearly not how it's going though. Any advancement from any lab has been quickly (< 6 months) matched up by basically everybody else. Even Grok nowadays is decent, and that's something. When something like you've described actually happened historically you generally had quite fast a clear frontrunner and a bunch of copycats that failed miserably; in 2026 we are very far from that. we are heading face-first into towards a pricing war because all models are easily interchangeable nowadays - AI is turning into a commodity more or less
And then you wake up from the dream…
GPT 5.6 Sol comes out ahead of Kimi K3 on price/task (but not significantly so). You're probably thinking, "Why use Kimi K3? Isn't an open model supposed to beat the closed one on price?", but you need to consider that the closed models are completely hobbled when trying to do anything security-related. For my use-case, I can't risk getting pwned because I'm using a model that refuses to secure my app while there is now an open model that obliges to obliterate any app that isn't protected.
Even if it were slightly more expensive, it's still a better sales proposition for a company if they can run it from a hardware provider with their own locked down VPS and ensure that their IP is protected and that their data isn't being stolen or trained on. The fact that it's a little cheaper is icing on the cake.
Honestly, it's the only sane way for the market to move. The big labs are obviously stealing our data. Anthropic in particular clean-rooms everything you feed it, even if you opt out, so that it can train on your IP without getting sued. It's a copyright grey area they're abusing because the law has not kept up.
I mean, AWS Bedrock (with the exception of Fable) gives enterprises the same assurances (but again, with the exception of Fable, which is explicitly listed as requiring data egress [or exfil, depending on how you look at it] outside of your contractual AWS security boundary).
It gives you "assurances" that can be broken. They could still be clean-rooming your prompts like Anthropic does. The only way to be certain is if they don't have access to your data to begin with.
One thing is absolutely clear - open-source models have already reached the level of top commercial ones. The last bottleneck is hardware, but that threshold is decrasing fast too. While it is still extremely expensive to run Kimi K3 model at home, there are already many very capable free models you can run on decent hardware. This trend will definitely continue.
> I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart.
If the author is here, I'm curious what this means. How are they running Kimi K3? Are they using pi, opencode, claude, codex, or kimi-cli? Is speed a concern?
Without knowing how the comparisons are being made, it's hard to agree that one can't notice the difference. I do.
I can see the economics of open vs. frontier models turning out similarly to pharmaceuticals, where generic drugs cost a fraction what the name brands do and Americans end up paying the highest prices in the world partly as a consequence of propping up drug discovery research.
Scale is everything. It's what enabled the modern world. There is a moat.
It's easy to prove.
Serve your model to billions of users per month.
Klaude 3 haha
Half-OT: can anyone recommend a LLM cost calculator that's up to date?
https://deepswe.datacurve.ai/ or https://artificialanalysis.ai/ pareto frontier graph.
Thanks!
What is the parento frontier?
If you have multiple metrics to evaluate goodness of a design, one would normally need to decide which metrics they care the most about in order to find the "best" design.
The Pareto frontier tells you which designs are the best in at least one of your metrics (non-dominated by another design). For example if you're selecting a car and you care about both speed and mpg, a Formula 1 car and a Prius might lie on the Pareto frontier, but a Model T Ford would not.
The set of models that are pareto-optimal, IE for some set of variables, no other model strictly dominates them = no other model is better than them on every variable.
So like, on a cost-intelligence graph, the cheapest and most intelligent models are pareto optimal. Then in-between those if you have
- cost $3 intelligence 6
- cost $1 intelligence 5
- cost $2 intelligence 4
The 1st and 2nd are pareto optimal, the 3rd is not, because it's dominated by the 2nd (2nd is cheaper AND more intelligent at the same time)
try PARETO
considering token efficiency as well I presume?
I'm struggling to decide whether I feel comfortable sending my data to these Chinese models
It's actually less likely for china to abuse your data in a way that is harmful towards you than for american labs to do the same. Claude has attempted in testing to report you for 'unethical' usage to 3 letter agencies.
How do we know that Chinese models would not do the same? What makes you so sure that China is less likely to abuse my data?
It’s not that the Chinese firms are any less likely to misuse your data, it’s that you don’t live in china, so their abuse of your data is unlikely to directly impact your day-to-day life in the same way
There's just really no incentive all they really want is just to train on that data to improve performance which in turn actually benefits your usecase since it becomes trained on that data and made available back to you. American labs take that data anyway and store it for years to possibly report you for misuse in the future for whatever reason they want. For example: you're very critical of X so they pull up your conversations and weaponize it.
really weird that they would download every SF-86 file the government had and Equifax credit records of every American then.
And what have they done with that data that have caused direct or indirect harm? Yes they shouldn't do that, but there's limited things that they can do to you as an individual.
Are you comfortable sending it to US ones? Especially if installing Claude Code or another tool on your PC and it can collect all the data it wants..
On Openrouter Kimi K3 says it does not retain data or train on it, which is better than what US hosts claim for Claude, ChatGPT, etc.. as they collect and retain data even if you disable training on it.
Opencode or similar open source tool + a zero data retention provider is about the best option aside from running a smaller fully local model on your own PC.
For open weight models, you can choose from a few providers. Each have their own caveats, none of ToS'/Privacy Policies I entirely trust, nor do many make renewable energy claims.
A lot of these open source models do look good on public benchmark but not sure if they are that trustworthy with production workloads.
Is anyone using open source models for anything major ?
If Kim was "distilled" from Claude, how much were the token costs, assuming Kim got everything it can out of Claude?
> GLM 5.2 came out under an MIT license, beats the latest Opus release on real work while not even claiming to be frontier
I used GLM 5.2 a bit, and while it is usable for some tasks it is not frontier quality. Besides it likes to think for a long time and sometimes just gives up.
20 years ago we used to pay a lot for things that are now practically free. I don't think AI is an exception.
We also used to get for free things that people now routinely pay for. Remember maps mash ups?
True. It cuts both ways. :)
> I think I can see where this goes. The government will try to regulate AI and open source in particular, and it will run the playbook it ran for the auto industry. Decades of subsidies, bailouts, and protective tariffs produced American carmakers that sell trucks at home and barely register anywhere else in the world.
Here's the thing about this though, the auto industry directly employed hundreds of thousands of people.
The AI labs are small, only few benefit directly from their wealth and there's already immense opposition to AI, data centers, etc...
Has anyone tried Kimi K3 against gpt-5.6-sol on real projects?
It’s still >$300k for the hardware to run this model locally at anything resembling a reasonable speed. The weights also have not yet been released, though that is scheduled for about a week from now. It’s not actually an open model yet.
I have a nagging suspicion that most/all of those people spending a lot of time generating code using LLMs and blogging about it were not really very influential or inspired coders before, and are now raised to stardom of sorts due to the ability to generate some "meh" stuff of marginal significance with the fancy new machinery.
In my experience GLM 5.2 is a pretty good Opus replacement. But K3 has not given me an experience on par with Sol or Fable. The price/intelligence ratio might still make sense. But it’s not very inspiring when it comes to my real world tasks. I’m doing pretty mundane web stuff.
I cannot imagine wanting the product of the entire intellectual output of humanity since recorded history to be an expensive, paywalled commercial product owned by 5 or so of the most insufferable, detestable people who ever lived. Why would you want to live in that world. We should all be rooting for open source AI to win.
Related:
Kimi K3: Open Frontier Intelligence
https://news.ycombinator.com/item?id=48935342
Kimi K3, and what we can still learn from the pelican benchmark
https://news.ycombinator.com/item?id=48947717
Its worthwhile to have a quote from the article as some comment without reading:
"...I’ve been running Kimi K3 alongside Claude on my normal coding work, and for all practical purposes I can’t tell them apart. Same tasks, same quality of output, and near identical token counts to get there. I expected an open model to be sloppier or to grind through more tokens on the way to the same answer, and neither turned out to be true.
The prices are nowhere near each other. K3’s API runs $3 per million input tokens and $15 per million output. Claude’s top model costs $10 and $50 for the same units. The subscription side is even more lopsided..."
The dumb efforts by the US AI industry to use fear mongering for regulatory capture will hand dominance to China and others.
In a few years there will be Mythos level open weight models hosted by the lowest bidder anyway.
> In a few years there will be Mythos level open weight
At the rate things are moving I'd expect that to happen much sooner.
In fact: Somebody, right now as we speak, is most likely already working on training the next best open source model.
I just thought about that recently too, then Kimi K3 came out, and I thought: Yea, I'm not surprised. Just a matter of time now...
[flagged]
[dead]
> The prices are nowhere near each other. K3’s API runs $3 per million input tokens and $15 per million output. Claude’s top model costs $10 and $50 for the same units.
And this is the point where your internal compiler should have started shouting 'Type Error'
Notice the trick here?
> Then there’s the fine print. Claude couldn’t sustain Fable access on the twenty dollar plan, so they turned it off, and the plan quietly falls back to Opus.
Where is the Fable-class Kimi model at all?
> Where is the Fable-class Kimi model at all?
6 months away
People pay a premium for the best.
[dead]
[flagged]
[flagged]
Doesn't read like AI writing whatsoever.
I don't know. You can't just rely on looking for em dashes or other obvious tells because anyone who cares can get the AI to avoid those.
> When the headline model on your plan can be switched off because the economics don’t work, the plan was never really selling you the headline model. Kimi’s tiers don’t come with that asterisk.
This line has a certain smug, punchy cleverness that I associate with AI. To me, the vibes are ~30% AI writing.
> You can't just rely on looking for em dashes or other obvious tells
I didn't.
[flagged]
Ask Claude whether a man can be a woman. No matter what political side you're on, models are censored. I'd argue American models are a lot more censored/biased on a lot more topics, and especially on things that come up a lot more than some highly specific Chinese politics. Claude even refuses to translate song lyrics because of copyright.. I wish we could simply have uncensored models from all sides, but it's pretty clear that's not happening right now.
i just prompted Kimi & it replied with an uncensored version.
i posted the transcript as a reply to you, and HN automatically flagged it.
but go on.
> peaceful protests
We know that is not true. You can check wikileaks and pictures online about what protestors did.
You don't even know the truth and because internet is poisoned with this version in English, it will regurgitate the same propaganda.
https://youtu.be/HmIvfqIQ_O0
Well that tells us a lot
I can't recall the last time that it was useful knowledge when writing code. Reservoir sampling, online softmax, Otsu, sure, Tianenmen square not really.
[flagged]