Is Opus 5.5 nerfed? Thinking patterns changed on November 3
Points and comments are a snapshot, not live.
A user claims Claude Opus 5.5's reasoning declined sharply on November 3, based on analysis of 1,400 sessions.
Shane Melaugh claims Claude Opus 5.5 was 'nerfed' on November 3, citing a Gemini 4 Argon analysis of 1,400 Claude Code sessions showing a sharp change in reasoning patterns. He also reports a nosedive in reasoning volume for 'high' mode, and a test where 'high' mode failed to fix a bug after 3 hours while 'xhigh' succeeded in minutes. He concludes the 'use high for everything' advice no longer applies.
What commenters are saying
Many commenters agree they noticed Opus 5.5 decline in quality, mistakes, and consistency. A common theory is that models are quietly downgraded before new releases to make them look stronger. Some defend against nerf claims, citing unseen performance variance and cognitive bias. Proponents of open-weight models argue they offer control and consistency that APIs cannot. One user reported a drop from ~85% to ~10% success rate in their Fable workflows two weeks ago.