کلوڈ Opus 5.5

Anthropic's model for long-running agentic coding and knowledge work. It uses always-on adaptive thinking, with medium as the direct API's default effort. The model accepts text, images and documents within a one-million-token context and supports up to 128,000 output tokens.

Anthropic presents Opus 5.5 as a model for sustained software engineering and knowledge work. The lab reports performance approaching Claude Fable 5.1 while reducing the cost of completing tasks compared with Opus 5. These are provider evaluations, whose results depend on the task and reasoning effort.

The release emphasizes follow-through: investigating a problem, checking evidence and continuing through longer assignments. Anthropic describes improvements in coding, research and communication, including clearer explanations. That combination suits work where an answer needs both investigation and a useful written result.

The announcement also describes limits and safeguards. Anthropic describes additional safeguards for higher-risk biology and cybersecurity requests. The published examples do not establish reliability for every application. Users should check consequential conclusions and validate generated code within their own projects.