AnthropicClaude Opus

Claude Opus 5.5

Claude Opus 5.5 is aimed at extended coding and knowledge work. It supports vision and computer-use workflows, but several important limits differ between ordinary requests and the batch service.

The larger output limit is not the default

The 300,000-token output option belongs to Message Batches and requires a beta header. Do not size an interactive request around that number. The regular maximum is lower.Claude Opus 5.5 reference

Short prompts do not qualify for caching

Opus 5.5 requires at least 512 prompt tokens for caching. The service also prices cache writes, reads and ordinary input separately.Claude Opus 5.5 reference

Make large code changes reviewable

Practical guidance. For a refactor, ask for the intended behavior, changed interfaces and verification evidence. Keep business requirements explicit. A model can produce a coherent explanation while still missing an edge case, so review the actual change against the requirement.

Specifications & access

Claude Opus 5.5 specifications
DetailValue and conditions
Launch date
AccessAPI
Model identifiersclaude-opus-5-5
Context window1,000,000 tokens

Context limit

Claude Opus 5.5 reference
Maximum output128,000 tokens

Ordinary maximum output

Claude Opus 5.5 reference
Maximum output300,000 tokens

Message Batches API with output-300k-2026-03-24 beta header

Claude Opus 5.5 reference
Input price4 USD per million tokens

Standard input rate

Claude Opus 5.5 reference
Output price20 USD per million tokens

Standard output rate

Claude Opus 5.5 reference
Thinking controlAlways on; adjust depth through effort rather than turning it off.Claude Opus 5.5 reference
Default effortmediumClaude Opus 5.5 reference
Minimum cacheable prompt512 tokensClaude Opus 5.5 reference

Related models

  • A model for coding, writing and agent tasks with configurable reasoning.

    Launch date API
  • A reasoning model for complex tasks and work that spans many steps.

    Launch date API