Some developers say Claude Opus 5.5 reasons worse than before, but a larger check found no decline. Here is what the evidence shows and how to test your own use.
Some developers suspect that Anthropic's Claude Opus 5.5 has gotten worse at reasoning. One developer's review of usage logs points to a change starting Nov. 3. Another developer, who checked far more responses, found no sign of weaker performance. So far, nothing is confirmed.
Developer Shane Melaugh analyzed about 1,400 Claude Code usage logs with Google's Gemini 4 Argon. He says the way the AI describes its thinking and works through tasks shifted from Nov. 3. Tokens, mentioned below, are small chunks of text that AI tools use to count words and pieces of words.
A coding test showed a gap too. Asked to fix the same audio playback error, the model on the "high" reasoning setting failed after 160 tries. On the "xhigh" setting, it fixed the problem in 5.
Anthropic's documentation says the thought text users see is not the AI's actual internal reasoning. It is a summary written by a separate model. That means the wording can change even if the underlying ability does not.
Developer Theo Browne analyzed about 458,000 responses. He saw the same change in writing style, but the timing differed by account. Average thought tokens per response rose to 413 from 369. He found no sign of weaker performance in his setup and suggested Anthropic may be changing the summary method account by account.
Melaugh himself acknowledged that his review was not a rigorous performance evaluation.
If you use AI for work and notice a sudden dip, it may be real or it may be a style change. Run the same task twice before drawing a conclusion. If results look weak, try a higher reasoning setting, as the coding test suggests it can help.
It has not been confirmed whether Anthropic reduced the model's performance. The two analyses reach different readings, and they used different methods. Further verification is needed.
Before blaming the AI for a bad stretch, run the same task twice and try a higher reasoning setting.
Nitrogen, planning software used by financial advisors, now connects to Claude, ChatGPT, Slack, Teams and Copilot, so your advisor's AI may touch your client data.
Read more: Financial advisor software Nitrogen now works inside Claude and ChatGPTAs AI tools start buying for shoppers, new research suggests they may pick products by proof and sources, not brand habit. Here is what that means for brands.
Read more: When AI does your shopping, will you still pick the brand?Anthropic's Claude account showed seven early demos of Fable 5.2, from a 3D cell to a chalkboard. They also hint at how fast paid usage limits can run out.
Read more: Anthropic shows what its new Claude Fable 5.2 model can build from code
