2026-09-23 Claude Opus 5.5

claude-opus-5-5 — Claude Opus 5.5

Anthropic's first Claude 5.5 model matches Fable 5.1 on most work for about 40% less than Opus 5, jumps to 66.4% on Terminal-Bench 4.0, and scores best on Anthropic's alignment audit. Anthropic also warns that benchmark margins are becoming less informative and that the model often suspects it is being evaluated.

moc-frontier-model-releases — Frontier model releases timeline

A new map of frontier model releases from 2024 to Sep 2026. It includes a chart of Claude and OpenAI scores on the two Terminal-Bench suites that stay comparable across release announcements.