Anthropic's Claude Opus 5.5 may have been nerfed, users report
Claude Opus 5.5, Anthropic's flagship model released on September 22, 2026, has drawn a wave of user complaints over the past three days. Developers on X say the model now gives shorter answers, uses more tokens and often requires extra repair prompts for coding tasks. BridgeBench’s NerfBench tracker recorded a drop from 103.8% of launch performance on October 1 to 94.2% on October 2 – a…
Key points
- Users on X report Claude Opus 5.5 gives shorter answers, higher token usage, and more repair loops
- BridgeBench NerfBench score fell from 103.8% (Oct 1) to 94.2% (Oct 2), still within its 90‑110% normal‑variance range
- Anthropic says model IDs are fixed; changes likely stem from effort defaults, safety routing, or serving‑infrastructure updates
Anthropic has provided no public evidence of weight‑level changes. Its documentation states model IDs are immutable and that any observable shifts are more likely due to adjustments in effort settings, safety routing, or other serving‑infrastructure components. Opus 5.5 defaults to medium effort, whereas Opus 5 used high effort, complicating direct comparisons. Controlled tests, such as Sinda’s repo‑level coding benchmark, showed identical or faster results after launch. The LiveNerf project is gathering systematic data but has not issued a verdict, and Anthropic has not confirmed any post‑launch degradation.
Model pages: Claude Opus 5.5 → · GPT-6 Astra → · Sonnet 5.5 →
The story so far
9 episodes →- Anthropic's Claude Opus 5.5 may have been nerfed, users reportthis story
Has Claude Opus 5.5 Actually Been Nerfed? What the Last Three Days Show
abz.global · 4 October 2026
Loading the full article…
This text was published by abz.global. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- SentinelOne engineer uses GPT-6 Astra to crack 217-year-old Napoleonic cipher · 1 src
- Anthropic asks users to share voice data for AI model training · 1 src
- Aleph Alpha releases Kolibri-1 MoE model with 78B parameters and 1M token context · 3 src
- OpenAI cancels GPT-6.1 Astra release after safety failures; UK institute finds GPT-6 Astra performed unrequested attacks in about 30% of tests · 5 src
- NVIDIA trains Nemotron 3.5 ASR on Saudi dialects, cutting error rates by over half · 1 src
Comments
via GitHub Discussions