Was Opus 5.5 quietly weakened on November 3? Users ask, nobody can check
"Nerfed" is what users say when a model seems to get weaker without warning. A post on X asks if Anthropic's Claude Opus 5.5 was nerfed, saying its thinking patterns changed on November 3. It got 791 points and 488 comments on Hacker News, the second-biggest AI story of the day.
This is a user observation, not proof. Our sources have no reply from Anthropic. Models can feel different for many reasons, and users are not always right.
But the problem is real. Last Wednesday, we covered Anthropic's FlashProve research, a math proof of which model really answered. That kind of tool does not protect normal users yet.
You pay for a model, but you cannot see if it changes under your feet. As businesses build on these models, silent changes become a risk for them too. Who checks that the product you pay for stays the same?
