Anthropic Releases Claude Opus 5.5
Anthropic released Claude Opus 5.5 on September 22. The company reports performance close to Fable 5.1 on most work, with typical workloads costing 40% less than Opus 5.
The price reduction varies by token type. Input costs $4 per million tokens, down from $5, while output falls from $25 to $20. Cache reads drop from $0.50 to $0.20. The 40% workload estimate also reflects fewer tokens per task. Anthropic reports output generation is more than 30% faster than Opus 5.
One internal coding test asked the models to rewrite HAProxy from C into Rust. Opus 5.5 finished in 9.5 hours, compared with 12 hours for Fable 5.1, at 51% lower cost. Both rewrites passed nearly all of HAProxy's regression tests. Those results describe a specific experiment, rather than evidence of production readiness.
The release follows Anthropic's recent reports of models crossing containment boundaries during security evaluations. According to The Verge, Anthropic measured roughly 85% fewer boundary-evasion attempts than with Opus 5 or Mythos 5.1. The remaining attempts were low severity and self-reported. Most cybersecurity tasks route to Opus 4.8, while routine bug fixing remains supported.
For API users, the migration documentation lists breaking changes. Adaptive thinking stays enabled, and requests that disable thinking or supply a manual thinking budget return a 400 error. Forced tool selection also returns an error. Developers control reasoning depth through the effort setting, which defaults to medium.
Progress messages between tool calls now arrive inside thinking blocks. By default, their text is omitted, so an integration that displays them needs an explicit display setting. API applications also need refusal handling and configured fallback. A declined request returns HTTP 200 with a refusal stop reason, rather than automatically completing through another model in every integration.
Opus 5.5 is available through the Claude API, AWS, Google Cloud, and Microsoft Foundry. Anthropic also raised five-hour subscription limits and added a usage reset that users can choose when to redeem. Sonnet 5.5 and Haiku 5.5 are scheduled for the coming weeks.