Anthropic announced Claude Opus 5.5 on 30 October as the first model in what it calls the Claude 5.5 family. According to the announcement, it performs at the level of Claude Fable 5.1 on most work, and it is the strongest model to date on Anthropic's automated behavioural audit, the alignment suite it runs across thousands of simulated scenarios. On vendor-run benchmarks, Opus 5.5 scores 66.4% on Terminal-Bench 4.0 for agentic coding, against 55.8% for Fable 5.1 and 52.3% for Opus 5; 54.4% on FrontierCode v1.1 (Main); and 57.8% on CursorBench 4.0. Anthropic says early testers reported a 680,000-line code migration finished in under a day, and that on a task to cut load times across every page of a web app the model succeeded 39 times out of 40 without changing the app's behaviour. It also says the model is more resistant to prompt injection than Opus 5 and less likely to take hard-to-reverse actions. Anthropic states the model was tested before release by external evaluators including Frontier Design and METR, with full evaluation details in the Opus 5.5 system card.
Our takeThe benchmark table is vendor-run, and Anthropic says so itself: at this capability level it calls benchmark margins an unreliable guide and says the real gap to Fable 5.1 is narrower than the scores imply. That caveat is the most useful part of the release. The claims a third party can actually check sit in the system card, where prompt-injection resistance and refusal of irreversible actions are measured against a fixed scenario suite rather than a leaderboard. Treat the coding percentages as directional until independent evals reproduce them.