Is Claude Opus 5.5 Nerfed?

An X post on Nov 7 by @shanemelaugh asked “is Opus 5.5 nerfed?” It said Gemini 4 Argon had analyzed 1,400 Claude Code sessions and found a sharp turn in how Claude Opus 5.5 reasons. Its chart dates the shift to Nov 3 at about 18:00 UTC, with the same model ID and the same Claude Code version. The chart says “the shift appeared mid-session across 7 active sessions”, and that visible summaries show “a behavior shift, not proof of changed model weights.” Anthropic's own dated records don't show a change to Opus 5.5 at that time. One Claude Code release came about 16.5 hours before that time. The next came about 6 hours after it.

Medium effort since launch

Medium has been the default effort of Claude Opus 5.5 since Anthropic launched it on Sep 22. The launch post reports its scores “at default effort (medium)”. Anthropic's effort docs say that most Claude models default to high. They add that “Claude Opus 5.5 and Claude Haiku 5.5 default to medium.” Claude Code added Opus 5.5 the same day in version 2.1.280. Its notes say new models like Opus 5.5 “start at their default until you pick a level”. Claude Code's model configuration docs still say that Opus 5.5 starts at medium.

Two Claude Code releases around Nov 3

Anthropic released Claude Code 2.1.317 on GitHub on Nov 3 at 01:25 UTC. One of its changes concerns subagents and teammates that run on a different model from the main conversation. They now use “that model's saved effort level or its default, instead of the session's effort level.” The same release fixed a project's effort setting being ignored in sessions that an app starts ahead of time.

Claude Code 2.1.318 followed on Nov 4 at 00:01 UTC. It removed an earlier fix for conversations started on version 2.1.286 or earlier. That fix had kept the model's earlier thinking when such a conversation was resumed. The release says “that thinking can now be dropped.” Several Claude Code releases in late October also changed how it handles effort. Among them are 2.1.314 on Oct 30 and 2.1.316 on Oct 31. The Oct 31 release added a setting for the effort level of subagents.

No Opus 5.5 change in the release notes

The Claude API release notes list entries on Nov 2, Nov 3 and Nov 5. None of them changes Opus 5.5. The Nov 3 entry adds Anthropic's text watermark to Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6. Since the launch, the notes mention Opus 5.5 only in entries about fast mode and Dreams. Anthropic's status page doesn't list an incident between Oct 31 and Nov 7.

Anthropic also publishes the system prompt that Opus 5.5 runs with in the Claude apps on the web, iOS and Android. That page lists two versions, dated Sep 22 and Oct 27, and none from November. The Oct 27 version updated the list of Anthropic's products and models and added guidance on several sensitive topics. It doesn't mention the effort setting or how the model shows its thinking. The page covers the Claude apps, not Claude Code.

Anthropic's records don't say what changed in the sessions the post describes.