Anthropic ships Claude Opus 5.5, 40% cheaper than Opus 5

Anthropic released Claude Opus 5.5 on September 22, cutting input token prices to $4 per million and output prices to $20, down from $5 and $25 on Opus 5. Cache reads drop to $0.20, a rate that matters for long-running agentic workflows.

The company's announcement claims a "40% cost reduction versus Opus 5" and "30% faster output generation than Opus 5," then lists benchmarks: 66.4% on Terminal-Bench 4.0, 54.4% on FrontierCode v1.1, 81.8% (partial) on OSWorld 2.0, and 67.7% with tools on Humanity's Last Exam.

The demonstrations lean on endurance. One tester completed a 680,000-line code migration in under a day. Another developer said the model "stayed on task for over 18 hours" on repository work. Web optimization tests hit 39 of 40 on load-time reductions, and merger evaluation tasks came in 50% cheaper.

On safety, Anthropic reports an "85% reduction in containment boundary circumvention attempts" and describes an anti-distillation "preserved thinking" safeguard, a cybersecurity fallback to Opus 4.8, and biology capabilities gated behind a Life Sciences Verification Program. Pre-release evaluation ran through METR and Frontier Design. Thinking mode cannot be disabled.

The model ships on AWS, Google Cloud, and Azure under the ID claude-opus-5-5, with EU AI Act watermarking included. Three researchers in our Who's Who directory posted the announcement on launch day.