Claude Opus 5.5 Review

Opus 5.5 is the first model in Anthropic's Claude 5.5 family, released on September 22, 2026, and the pitch is blunt: it does the work of Claude Fable 5.1 on most tasks and costs 40 percent less to run than Opus 5. Anthropic's own table backs the first half. On Terminal-Bench 4.0 it scores 66.4 percent, against 55.8 for Fable 5.1 and 57.9 for GPT-6 Astra, and it also leads FrontierCode, CursorBench and the GDPval-AA knowledge work ranking, with 67.7 percent on Humanity's Last Exam using tools. Anthropic adds its own caveat, that margins at this level overstate real differences and the gap to Fable 5.1 is narrower in use than the scores suggest, and Google's table for Gemini 4 Argon, announced a week later, has Argon ahead on several knowledge work and coding tests while Opus 5.5 keeps the Terminal-Bench lead.

Where it earns the top grade is long, sprawling work. In Anthropic's test it translated HAProxy from C into Rust, passing nearly all of the project's own regression tests, in 9.5 hours against 12 for Fable 5.1 and at 51 percent less cost, and an early tester audited and fixed a 200,000 line codebase in under three hours where Opus 5 took more than twenty. The pricing makes that an everyday tool rather than a special occasion: $4 in and $20 out per million tokens, cache reads at 20 cents, output more than 30 percent faster than Opus 5, and fewer tokens spent per task. Unlike Fable, it is offered under zero data retention, and it shipped on AWS, Google Cloud and Azure on day one.

The costs are the ones the Fable line taught everyone to expect. Thinking can no longer be switched off, only turned down. It carries safeguards like Fable 5.1's, so when a cybersecurity or biology request trips them the work falls to an older model, Opus 4.8 or Opus 5, unless your organization is enrolled in one of Anthropic's verification programs, and Anthropic's own benchmark notes say that routing likely cost it points. For high volume, low stakes jobs it is still more model than you need. For everything hard it is the best value at the frontier and the new default, and it is an S.