Only if you buy their math, which basically is "hey, if we don't do any training we can actually make a profit".
The problem with that math is that if they don't do any training they would be out of the market in 12 months, they're only relevant ("profitable") precisely because they trained the current reference SOTA model.
They can't just release Mythos and sit on top of it forever, competition is catching up fast and people expect a new more powerful model every 6 months.
Are they quietly compacting context to reduce kv cache usage, before the actual compaction? Like there’s a slider for how much to compress it, and that’s never revealed to us?
You have to learn to think like a drug dealer. The first hit is always free.
Companies and developers are growing more and more dependent on coding agents. Eventually, the owners of the AI will be able to charge whatever they want. What are you going to do? Go back to coding by hand? Do you even remember how?
Can you please not break the site guidelines like this? They include:
"Don't be snarky."
"Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something."
"Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith."
"Don't be curmudgeonly. Thoughtful criticism is fine, but please don't be rigidly or generically negative."
https://news.ycombinator.com/newsguidelines.html
Is there any evidence that they nerf models? Anthropic is set to mark a profit Q2 2026 (which is actually not ideal), but there is profit.
Only if you buy their math, which basically is "hey, if we don't do any training we can actually make a profit".
The problem with that math is that if they don't do any training they would be out of the market in 12 months, they're only relevant ("profitable") precisely because they trained the current reference SOTA model.
They can't just release Mythos and sit on top of it forever, competition is catching up fast and people expect a new more powerful model every 6 months.
There are LLM performance trackers in the wild, for instance https://marginlab.ai
You may notice that the performance of the old model tends to decline before each new model release.
Wasn't that non-GAAP profit?
How do they “nerf the models”?
Are they quietly compacting context to reduce kv cache usage, before the actual compaction? Like there’s a slider for how much to compress it, and that’s never revealed to us?
I suspect they quantize them, reduce thinking budgets, batch more requests, or all of the above.
There's also lowering the number of experts you run in MoE models.
> there is no 'profit' step.
You have to learn to think like a drug dealer. The first hit is always free.
Companies and developers are growing more and more dependent on coding agents. Eventually, the owners of the AI will be able to charge whatever they want. What are you going to do? Go back to coding by hand? Do you even remember how?