It's weird because two days after Jev was released there were a dozen decision models, a week later there are several dozen, mostly open source, OpenAI's own Decisions API [1] beats it, and you can easily finetune your own [2]. But as others have pointed out, this doesn't matter.
EDIT: As I wrote this Microsoft just released their own Decision-1 model [3].
[1] https://developers.openai.com/api/docs/guides/decisions
[2] https://unsloth.ai/docs/basics/train-your-own-decision-model...
[3] https://commandline.microsoft.com/microsoft-decision-1-model...
Decision models have the potential to have an even larger impact on the Real World than LLMs have to this point (which is obviously quite large). But the model itself matters less than the product experiences you build around the model, and its very likely that the incumbent labs are treating the area as something more like "oh yeah I guess we can ship that and then forget about it" rather than investing in what building business processes on decision models looks like. Unlike full language models, I don't think the primary business of Typesafe will be serving Jev at API pricing; it'll look a lot more like putting Jev at the center of a much more expensive suite of software.
There's the potential for an inverse LLM play. In contrast with LLMs, all that seems to matter is the model, and the products the labs build around the models are all really samey and boring; the same left panel list of agents, main view agent conversation, right hand extra context, and we're now in the era of everyone creating the same cutesey furry friend on top of all this tech.
> But the model itself matters less than the product experiences you build around the model
This is a very important insight. And it applies to LLMs as well. Very few people were impressed with the capabilities of GPT 3, it was mostly a techie novelty
But then when they added chat on top of gpt 3.5, all of a sudden it was a huge hit. Sure there were improvements in the model from 3 to 3.5, but the biggest impact was from the chat experience
Conversely, when they created Eliza, a basic chatbot more than 50 years ago, people even got addicted to it, despite it’s ai model being something super rudimentary and basic compared to what we have now. The model capabilities didn’t matter as much as the experience the chat created
> Decision models have the potential to have an even larger impact on the Real World than LLMs have to this point
Why?
If youre the one or have spoken to someone implementing "AI solutions" inside large companies recently, a decent chunk of it is soft policy enforcement with very basic context. They moved from gemini 2.5 flash lite type models to jev. Which is also why I found the price comparisons to "GPT Astra" on Twitter rather funny.
Yes, the best way to think of a general classifier like this is like a smart switch statement. Essentially a "JEV" like thing becomes a sort of programming primitive. Once you see it, it's hard to not get excited.
But even if others surpass them and make better solutions, the fact that nobody was able to see it before typesafe is a testament of what they might be able to come up with next.
I sound like a fanboy but I swear I 'm unaffiliated with typesafe. I was building my own version of this way before they announced JEV (mine was ALE and it was mentioned here on HN for a bit), in use for VR gaming (so one can give commands to NPCs with voice and supports multiple commands in sequence in a single pass), but I missed the "killer usecase" of being a new primitive, like everyone else.
TLDR: There's a lot of value in thinking ahead and seeing the future. The clones are nice and exciting but they give me "I could have built this first, yes but you didn't" vibes. I hope they manage to keep it up and push the space forward again
> [...]the fact that nobody was able to see it before typesafe is a testament of what they might be able to come up with next.
Counterpoint: there are a lot of one-hit wonders, and they vastly outnumber the idea-factory people. This is not to minimize those people, a single idea can be very successful (see Zuckerberg), but it doesn't mean your subsequent ideas will also be great (see Zuckerberg)
I remember your blog post! Thanks for writing it, was pretty cool and a practical application.
Here if anyone is interested: https://pantel.is/projects/ai-gaming-companion/
Honestly, I don't feel the least bit of excitement here and I'm normally enthusiastic about AI.
Yes, this generates probabilities over a set of given options instead of the whole token vocabulary.
I don't see what's so exciting about it, compared to regular LLMs which support structured output options in the API.
You’re correct, it’s a great solution for a very narrow set of high volume classification needs that require very low latency. But that’s it.
I keep evaluating Jev for my product because of the hype, but the reality is that for my tasks, Luna is more accurate, only a little more expensive, and the latency doesn’t matter. I’d rather spend the extra $50 / month or whatever than have to shoehorn in another API and provider, and also lose the ability to change reasoning level and get reasoning summaries for eval purposes.
>> TLDR: There's a lot of value in thinking ahead and seeing the future. The clones are nice and exciting but they give me "I could have built this first, yes but you didn't" vibes. I hope they manage to keep it up and push the space forward again
This all sounds intelligent and likely, and yet we can come up with countless counter examples where the first mover is not the big winner, and nobody cares about who did it first. There is typically way more value in nailing the execution of a big idea someone else came up with, rather than "seeing the future".
> Unlike full language models, I don't think the primary business of Typesafe will be serving Jev at API pricing; it'll look a lot more like putting Jev at the center of a much more expensive suite of software.
Precisely this. Should be top comment.
Also, the model moat is understated as training data for these purposes also accrues to the winner, which due to the first mover advantage as well as the distribution advantage you speak of, is typesafe. In contrast to relatively open coding data. Openai anthropic also have that, but like you say its a different business.
The are still just 1tok output of pretty standard llms just along with the logprobs converted to some json
At least in theory (TypeSafe has been pretty close-lipped about the details so this might just be hot air, and I think the evidence is a bit spotty) this is false, since they use a different reinforcement training method.
If the word “calibration” in probability doesn’t mean anything to you, the difference isn’t very apparent, but that doesn’t mean it doesn’t exist.
[dead]
Yeah, agreed. A model or primitive on its own has no moat and frankly limited value. The paradigm behind "System One" models on the other hand is potentially huge.
https://seldon-ai.com/blog/fronter-llms-are-semantic-interpr...
Skimming the post, it seems to argue for reconstructing the very rigidity that LLMs let us escape, and that very aspect of LLMs is what made them useful and explode in popularity so much.
But Jev established the branding and investors are betting that Jev will be acquired by one of the big labs soon - and if they aren't, the money itself can create a positive outcome by allowing Jev to hire incredible talent and scale the company rapidly.
Rapidly? Their 2 years of stealth was replicated in 2 weeks.
I give it to them for creating the hype (good marketing), and for making a useful classifier. Not sure what they would scale rapidly though.
I feel like half the game now is marketing though, so I can see why they'd be attractive to an investor. Maybe if they scale they can come up with something.
Always has been imo.
You can replicate any fitness app in no time and you will make close to $0. Brand recognition matters a lot.
Yeah but there are also network effects of using fitness apps. Every engineer I know switches easily between Codex and Claude Code when one gets better than the other.
Jev has been around for a couple weeks. Cost and performance matter more than anything. Staying with an existing provider (the # of people choosing Jev without already having a frontier API key is probably zero?) is way easier than this.
What the hell are we even talking about.
100%. Product finesse + Marketing is now the "moat".
It’s the classic SV flip. You scale your investors, your executive team, your sales people, hire a bunch of engineering you don’t need, then sell the company. The company’s product doesn’t matter, the company is the product.
It’s like openclaw. There was a bunch of technically better ones that came along afterwards, but nobody remembers what any of them were called.
You say this, while Hermes Agent has been at top of the openrouter.ai leaderboard for several months and currently has 3X the token usage of OpenClaw.
What? OpenClaw has been completely replaced by Hermes and others in the discourse
And Muse is what normal people use instead
> Their 2 years of stealth was replicated in 2 weeks.
Actually they stolen the idea from a paper.
So did Oracle with relational databases by that logic
Either the investors know something we don't, or the market is irrational.
[dead]
Most of them appear to be small LLM’s fine tuned for the role.
That’s a different set of properties in terms of size, cost, and latency. Jev (apparently, not like I’ve seen its insides) is extremely cheap, extremely fast, doesn’t cost any output tokens as it speaks the output natively, can’t get the output wrong because it speaks the format natively, and (presumably based on the docs), the context is separate from the question, meaning it should be immune (or at least highly resistant) to prompt injection attacks.
It’s not just about the accuracy of the result, it’s a collection of all the properties that make Jev interesting.
Jev took years to develop, I strongly doubt that a copycat that was put together within days after Jev’s release will be able to match it on a sun of its properties. Even if fine tuned LLMs can outperform it on raw accuracy.
Jev is something your favorite LLM could zero-shot months ago, if you pointed it to the right arXiv paper (some of which are linked in this thread).
That probably explains why there were so many competitors around withing days of the Jev announcement. They are not starting with a moat, and there doesn't seem to be any moat in sight. Just buzzword recognition because everything is comparing to "jev".
There will always only be a very small percentage of people who want to discover and build things, even with llms, its a very small subset of people though, and most people want off the shelf solutions.
Also Jevs purpose isnt to become its own thing. It will get aquired in 18 months by one of Andressen Horowitz's incestuous circle of companies and everyone will make money, and the person who buys it wont necessarily care if Jev itself makes them a ton of money. They're just passing chips around the table.
Jev did not take years to develop. What it does was published in arxiv back in 2025. TypeSafe just marketed it.
What specific arxiv paper are you referencing here?
https://arxiv.org/abs/2503.23303
and https://arxiv.org/abs/2510.01237
Yeah, I figured it was gonna be this one.
"SalesRLAgent: A Reinforcement Learning Approach for Real-Time Sales Conversion Prediction and Optimization"
Jev is a general-purpose thing. That is a specific-purpose thing. General-purpose thing is not the same as specific-purpose thing. What makes people think these are the same thing? I don't get it.
What do you mean? Jev is trivially different from what is described in this paper.
Bullshit. Below is copypaste from the paper in the section that outlines the "key contributions" of the paper. As you can see, it is focused on one specific problem: predicting sales conversions. So if you were to take this system and use it for some other task ("evaluate customer mood" for example), it would not work. Because, again, it is not describing a general purpose solution. It is describing a solution that is specific to one problem: sales conversions.
Copypasta:
• A reinforcement learning architecture specifically designed for sales conversation analysis and conversion prediction
• A synthetic data generation pipeline leveraging GPT-4O to create diverse and realistic sales conversations
• Novel state representation techniques using Azure OpenAI embeddings (3072 dimensions) with sales-specific features
• A meta-learning approach enabling the system to express confidence in its predictions based on conversation similarity to training data
• Integration mechanisms providing real-time guidance within existing sales platforms
• Extensive comparative evaluation demonstrating significant performance improvements over LLM-based approaches
> Novel state representation techniques using Azure OpenAI embeddings (3072 dimensions) with sales-specific features
Jev is basically the embeddings side of an LLM. Yes, it's a good idea, but the moat is non-existent.
No. It can't simultaneously be both general purpose and having task-specific embeddings.
This is a fine-tuned model. The author even states that the model is competitive with Jev only if fine-tuned on the evaluation at hand.
Literally misses the point of Jev, which you don't need to fine-tune to get accuracy nor - and no other model has this - some sort of out of sample calibration
> I strongly doubt that a copycat that was put together within days after Jev’s release will be able to match it on a sun of its properties
But why? If the simplest way to achieve Jev's capabilities (accuracy, cost, latency) is by fine tuning a small model, what makes you think that this isn't exactly what Typesafe did?
And even if they did something different - what makes you think it was a good idea in the first place, given how easy their results were replicated without any "secret sauce"?
My point is that it’s not replicated. You replicate the accuracy, but not the other properties. The Jev-competitors only proved that you can get or beat the accuracy, nothing about the other properties. Especially the “zero hallucination” output and the (if it works how the documentation make it sound) prompt injection resistant architecture. You can’t get that with a fine tuned LLM.
> zero hallucination
Plenty has been said about this claim. If you're still falling for this, I feel sorry for you.
If you remove wheels from your car, your car will get a "no speeding ticket" property, and yet there's nothing exciting about it.
> You can’t get that with a fine tuned LLM
Of course you can. All these claims are nothing but marketing.
I understand most major model providers support passing a JSON schema that is strictly followed in the output, accomplishing the same 'zero hallucination' and prompt injection resistance.
The only difference I am aware of is that probabilities are better calibrated with these decision models compared to regular LLMs which can output hallucinated numbers where your schema allows a number.
Jev outperforms clef and OpenAI decisions on most of my tasks, outside of when I need multi modal image going into it. Jev is also cheaper but costs are minimal overall.
I just dont understand why anyone needs a general purpose classifier. Just build a classifier for your specific use cases.
Fortunately Jev is cheap, so I dont think it matters too much, but I think its robbing people of the opportunity to learn and implement this themselves.
Also, I dont really want 3 companies responsible for censorship/classification.
I see Typeface AI have 30-40 people working for them on LinkedIn (some are VC advisors / board members), a lot of them being engineers. Their openings suggest a strong developer market focus (to begin with). I don't necessarily see a lot of people at the company with enterprise sales channel experience, but they already claim to have several Fortune 500 companies as clients (maybe the VCs are helping there or there is natural dev-driven traction).
So they clearly have a product, a strategy around it and perhaps the compliance scaffolding (SOC2 Type II etc) that may be needed before actually being able to charge money for it. They also have the right brand names associated with the founding team. Execution, so far, seems good enough to create a splash, at least.
As an investor, the question(s) to ask is (in my view): "How do they make money? Will that way to make money survive?". The answer to the first: selling input tokens and perhaps subscriptions/credits eventually.The answer to the second: "Yes, but with the risk of unit revenues declining faster than their unit costs". How can they mitigate this problem: by being big (scale / mindshare etc) so that their unit costs (including for customer acq) fall faster than their unit revenues will - I believe that is the question most AI companies are trying to tackle these days. Any new competitor will have to tackle basic fixed costs (of getting started) first before even getting to the stage of having the luxury of worrying about unit-economics.
So yes, they might eventually be competed away but whoever is in their team is trying hard to make a useful product/ecosystem and that should be applauded, not ridiculed with "it's all marketing". This is way more than a simple github/huggingface-based open-source replica solution can hope to achieve without institutional backing (either big-tech or system-integrators).
What should rightly be questioned, of course, are the valuations the VCs are providing to them in hopes of passing this hot potato to a willing buyer (say a hardware maker like NVidia) - the incentives there are very well defined and depend very much on perceived TAM (which lately is on very shaky ground given how far token pricing has fallen causing, among other things, OpenAI to "miss" on the market's expectations for annualized revenues, even before they're listed!) [1]
[1]: https://www.ft.com/content/b66a9858-f8fb-46cb-b506-44bfe26fc...
An Enterprise client at that size can just be someone at an f500 put a credit card in
> OpenAI's own Decisions API [1] beats it
Have you heard that from a different source than OpenAI? From what I'd heard other models haven't gotten close, and the open source ones are like running gemma4 E2B against Opus 5.5- sure, the API calls go in and are returned the same but the quality isn't close.
> you can easily finetune your own
no, you can't, and it's unclear why you would think this.
you can easily finetune your own*
*if you have a sufficiently sized and quality dataset for the specific classifications you're targeting
And even if you do have that, you haven't made your own Jev, because Jev is a general-purpose thing, whereas what you have built is a specific-purpose thing.
Yup, I also released an open source classifiers tool, Jeffy. It comes with 68 pre trained classifiers which run and train on CPU alone. They run locally and are faster than Jev/Laya/Decisions. And they can do things like label email, all the way to even playing Doom
* https://jeffyclassify.com/
* https://playground.jeffyclassify.com/#doom
* https://github.com/nicobrenner/jeffy
Isn’t the OpenAI decisions API basically just Luna cosplaying a decisions model and pretending the confidence score isn’t just a hallucination?
And what do you think Jev confidence score is?
Here's a hint: confidence is not generated by a model.
Maybe I'm missing something, but why couldn't it be generated by the model? In older classification tasks with transformers like BERT, you could absolutely obtain a confidence score.
Jev API returns both confidence and probabilities.
But confidence value is just a function applied to probabilities. It is not coming from the model, and it carries no additional information.
It is documented btw, and yet you will see plenty of claims that Jev is better than LLM because it returns both.
Thanks, fixed my understanding!
Do you think though that Luna being a model post-trained for chat produces over-confidence in logprobs?
Yeah, but I wouldn't be surprised OpenAI's decision API is a post-trained Luna with confidence calibration.
Typesafe claims that Jev is calibrated, but there are plenty of examples where it completely fails (predicting die roll being the most obvious one).
Unfortunately calibration is hard to benchmark.
The dice roll prediction is about the way the prompt is setup misunderstanding how Jev works (they treat the confidence score as a probability score, which it isn't).
If you instead give it a list of probability for each number and ask it whats the probability of each number, the result will be accurate.
What’s the difference?
Counterpoint: my work has already allowed us to call and test Jev. Those others? Who knows when, if ever.
In my experience with OpenAI's decisions endpoint, it tends to return either 0 or 1 and doesn't return middle confidence levels very much at all. Would be interested to hear if others have experienced the same.
I honestly was worried for them. Now, even with all the clones, they generally still come up on top in price and latency, but "good enough" is often sufficient.
Guess the (investment) market has spoken.
Hm? Jev's actual (network) latency is not that great and even small local models are doing far better latency-wise.
The Microsoft article GP linked even shows the MS model having 95ms latency.
Even on price Jev being matched (the same MS model is "Input tokens cost $0.042 USD per million tokens. Output tokens are free.", same as Jev).
Cloudflare's Clef-flash model is actually slightly cheaper: "$0.038 in / $0 out per 1M" too.
Jev is being matched or exceeded in performance and price within a month of them going public.
You are right in terms of how fast competition created alternatives.
But, for OpenAI this is not a primary business, for open source models as well, so they will not be chasing the market and customers to buy their product and promise them to maintain it.
TypeSafe will do all this, they will try to understand your use cases and then solve your pain point, while others are providing raw material.
A major VC could type safe ai money, then head to a larger AI company looking to raise their series E+ and demand they acquire typesafe as part of their funding allotment.
such an arrangement can end up beneficial to the VC firm
OpenAI's "Decisions" library has this in requirements:
To run the SDK examples below, use these OpenAI SDK versions or later: Python 3.26.0,
I thought Pythin 3.15.0 just came out, 3.26.0 must be really far off?
That is their Python SDK version, not Python version
> OpenAI's own Decisions API [1] beats it
Jev is 42$/B but OpenAI is 100$/B token.
I mean I'm already ditching my Apple stocks because soon AI will be able to replicate iOS and MacOS.
Investments aren’t made because the product is amazing, they’re made because there’s a compelling exit scenario. Engineers don’t want to hear this but more generally, the critical success factors for a business aren’t product or engineering they’re relationships i.e. sales and team dynamics. If technical excellence dictated business outcomes in tech Salesforce wouldn’t exist for example.
You are assuming that the VCs have done their due diligence. For a "hot" company like Typesafe AI, most likely little due diligence was done. That's the way it's played.
"It doesn't matter"
By all means, become an A16Z LP.
Nothing beats nepo, brother.