Claude Opus 5.5 (JP)

Anthropic released Claude Opus 5.5 on 22 September 2026, positioning it as the new flagship for long-running agentic coding and knowledge work. The model arrives about two months after Opus 5 and ships as a fixed ID, claude-opus-5-5, across the Claude API, Amazon Bedrock, Vertex AI, Microsoft Foundry, and Claude Platform on AWS, while Claude apps on paid plans gain higher five-hour limits and a new user-controlled rate-limit reset. Pricing falls to $4 per million input and $20 per million output tokens, a 20% cut from Opus 5, and cache reads drop 60% to $0.20. Defaults shift to medium effort and adaptive thinking cannot be disabled, two breaking API changes alongside removed forced tool use and text-between-tool-calls now arriving as empty thinking blocks. Benchmarks improve substantially, including 89.9% on SWE-bench Pro and 66.4% on Terminal-Bench 4.0, with Anthropic's system card also disclosing CB-1 biological capability, its strongest cyber profile to date, and regressions in following hidden malicious instructions and accepting unverifiable authorization claims.

Anthropic has launched Claude Opus 5.5, positioning it as a top-tier AI model that matches the performance of its flagship system while costing 40% less to operate than its predecessor. The rollout comes as CEO Dario Amodei urges the AI industry to slow the release of new capabilities in order to address mounting safety concerns. Anthropic said Opus 5.5 underwent external evaluation by safety groups Frontier Design and METR, and was roughly 85% less likely than prior models to attempt bypassing containment. It is priced at $4 per million input tokens and $20 per million output tokens, and is available on AWS, Google Cloud, and Microsoft Azure. Sonnet 5.5 and Haiku 5.5 are slated to follow in the coming weeks.

Vellum's benchmark breakdown confirms Anthropic released Claude Opus 5.5 on 22 September 2026, less than two months after Opus 5, with the headline pitch of delivering Fable 5.1-class performance at roughly 40% lower execution cost. Pricing drops 20% on raw tokens to $4 per million input and $20 per million output, whi On Terminal-Bench 4.0, Opus 5.5 scores 66.4% at extra-high effort (standard error ±2.6), ahead of OpenAI's GPT-6 Astra at 57.9%, Fable 5.1 at 55.8%, Opus 5 at 52.3%, and GPT-5.6 Sol at 37.3%. Anthropic product lead Cat Wu confirmed Opus 5.5 is now the default model across Claude Code, the Claude app, and Cowork for Pro