1,400 Claude Code sessions suggest Opus 5.5's reasoning shifted abruptly on Nov 3
churchkey · x · 2026-11-08
A developer analyzed 1,400 of his own Claude Code sessions with Gemini and found Claude Opus 5.5's reasoning behavior changed sharply around 18:00 UTC on November 3 — same model name, same CLI version.
Key data from the charts:
- Visible thinking blocks starting with "I'm..." jumped from 25% to 61.3%
- Use of "I should / I need to" dropped from 55% to 10.9%
- Em-dash usage fell to 6.5%
The shift appeared mid-session across 7 active sessions; the author notes this shows a behavior change, not proof the model weights were swapped.
More from coding & agent
- Emergent UI Downside: Colleagues Now Expect You to Read Every AI-Generated Interface — sloppenheimer · 2026-11-08
- Dev Open-Sources a Three Kingdoms SLG Where Claude Plays Everything via MCP — Regular_Bison_2175 · 2026-11-08
- Meta Muse mobile review: persistent VM, editable memory files, and async approvals make it a control plane — Acceptable-Web-6296 · 2026-11-08
- Interp researcher: AI coders dismiss accumulated engineering norms at their peril — voooooogel · 2026-11-08
- Full Grok thermal printer video hits YouTube, edited with Codex — billyjhowell · 2026-11-08
- Recreating Seedance-Style AI Video with Blender Blockouts + MiniMax H3 Ref2Vid — Time-Ad-7720 · 2026-11-08