User reports on Claude Opus 5.5 performance degradation
User reports of performance degradation
Recent discussions on Hacker News indicate that a segment of users perceives a decline in the performance of Claude Opus 5.5. Reports describe the model as having difficulty following instructions, exhibiting poor attention spans, and demonstrating increased verbosity. Some users have noted that these issues appear to have intensified over the last few weeks, with others reporting sudden changes in behavior within the last 24 to 48 hours.
Common symptoms of model behavior
Users reporting these issues highlight several specific behavioral patterns:
- Context Rot: Some users report that the model struggles with long-running sessions, eventually "forgetting" previous instructions or configuration steps it previously provided. This is described by some as a form of "context rot" where the model becomes disinterested or confused after multiple turns.
- Instruction Following: Users have noted that the model is increasingly likely to ignore specific constraints or miss details in complex prompts.
- Verbosity and Token Consumption: There are reports of the model becoming excessively verbose, leading to rapid depletion of usage quotas without a corresponding increase in output quality.
- A/B Testing: Some community members suggest that these variations in performance may be the result of A/B testing or silent model updates, which could explain why some users experience significant degradation while others report no change.
Community perspectives on model reliability
While many users report frustration, the community remains divided on whether the model has objectively "gotten dumber" or if these experiences are due to other factors.
- Benchmarking vs. Anecdote: Some users point to automated benchmarking efforts, such as those found on platforms like BridgeBench, to argue that the model's performance remains stable.
- Alternative Models: In response to these issues, some power users are exploring alternatives, including older versions of the model (such as Opus 4.8) or migrating to uncensored models like the "Abliterated" 70B/120B series to avoid perceived "shock collars" or restrictive guardrails.
- Speculation on Future Releases: A recurring theme in the discussion is the speculation that performance shifts may be precursors to the release of a new model version, colloquially referred to by some as "Fable 5.5" or similar variations.
"If I let Opus 5.5 run for too many turns... it starts exhibiting signs of human dementia. It will forget it told me to run something and will then accuse me of misconfiguring something that it literally gave me a script to run." — Hacker News user Bender
"I think something has changed, maybe about two weeks ago, because there’s this fixation from the model where if you ever need to make a correction, it simply becomes disinterested with your instructions." — Hacker News user itopaloglu83