Boris Cherny No it doesn’t feel more intelligent, and some benchmarks even show it objectively isn’t. It’s more expensive to run (tokens wise) which means more money for you and less things done for us. If I’m wrong, may Anthropic increase our usage limits by at least 25% to match the token

Boris Cherny Feeling that ~35% tokenizer tax already. Hit my 5-hour Opus 4.7 limit in under an hour on the $200 plan. 🙃

Boris Cherny This is embarrassing levels of bullshit. 4.7 is not more agentic or precise it's a downgrade in every metric. Shorter thinking doesn't follow workflow prompts at all and Anthropic quietly made it 59 percent more expensive via tokenizer swap with 40 percent less effective context.

Boris Cherny co-work is completely broken after updating to 4.7

Boris Cherny It’s shallower, LESS precise, refuses to think, struggles at long contexts. It is a regression in every meaningful domain.

Boris Cherny Deeper reasoning = Deeper spending on tokens. At this rate you'd hit the limit before finishing a real task, which kind of defeats the whole point of a more capable model.

Boris Cherny I hate to be one of those guys, but 4.7 has not been a good experience. If you profile my account for F-bombs, you will see a huge spike starting mid-day today, at a sustained level well above whatever that baseline is. I don't know what you guys did to this thing, but good god.

Boris Cherny Sorry but if you have to "take a few days to learn how to work with it effectively" then the release is a fail. It should be as easy as 4.5 to 4.6 was to use. No extra time or learning curve required.

x.com/i/article/2044…

Boris Cherny So much bullshit. Bro, go to Reddit. Read yourself the total fuckup you guys did. Opus 4.6 was fucking good and now you ruined it. It's lobotomized. And Opus 4.7 does not follow instructions, questions everything and can't even use available tools unless called out. You can write

Boris Cherny Opus 4.7 feels less innovative, more rigid, the experience is more like hand holding a junior dev. It SHOULD feel like a senior engineer who is given a task and can reason about it without needing so much detail.

Opus 4.7 just forced me to learn something really interesting: Google Gemini has gotten way better at coding. 4.7 completely broke my workflow. Literally the singular worst experience I’ve had with any AI, Paul Jankura. Announcing a better model (Mythos) that customers can’t touch,