Please post benchmarks if you are going to claim Claude got secretly downgraded
retained text
I keep seeing so many posts swearing up and down that Claude got worse because XYZ. “I have been using Claude for months and I swear that it has gotten way worse lately”. 0 benchmarks, 0 evidence. Just the “I swear”. And I get that 100% this could be very true. But what is also true is that everytime we get a new model the more responsibility and delegation we give to our agent. This is almost like the inverse jevons paradox. The better the AI gets, the more open ended and trusting your prompts will be. As such, you are going to notice more errors by nature. When a model first releases people will prompt it normally. But then they will notice in what aspects it improved in and so they will unconsciously limit test it more and more with more open ended prompts and less code review. So I would love to see actual benchmarks from all these people who are claiming that Opus 5.5 got downgraded because of X conspiracy. I am not defending Anthropic in any way. I just would really like to see proof.
document fetched 2026-11-07 00:38 · canonical https://reddit.com/r/ClaudeCode/comments/1xpqs9t/please_post_benchmarks_if_you_are_going_to_claim