Back to changelog

Claude Opus 5.5

Anthropic's Claude Opus 5.5 is live on the gateway with a 1M context window, always-on adaptive thinking, and $4 / $20 per 1M tokens — 20% below Claude Opus 5 — with cache reads at $0.20.

Anthropic's logo on a glowing chip surrounded by a terminal window, a brain, an hourglass, a stack of documents, and a coin on a circuit board

Opus-class agentic runs are where most of a coding budget goes: long sessions, big prefixes re-read on every turn, and thinking on every step. Claude Opus 5.5 brings the same 1M context and 128K output ceiling as Claude Opus 5 at $4 / $20 per 1M tokens instead of $5 / $25, with cache reads at $0.20 and stronger long-running agentic coding, code review, and knowledge work.

Claude Opus 5.5

1anthropic/claude-opus-5-5
Spec Value
Context 1,000,000 tokens
Max output 128K tokens
Input / output $4 / $20 per 1M tokens
Cache read $0.20 per 1M tokens
Cache write $5 (5-minute) / $8 (1-hour) per 1M tokens, 512-token cache minimum
Thinking Adaptive, always on — steer it with reasoning_effort, low to max
Modalities Text and images in, text out; tool use, JSON schema output, and web search

Before you move a pinned claude-opus-5 workload over, note what changed:

  • The default effort is medium, not high. A request that omits reasoning_effort thinks less than it did on Claude Opus 5, and at a given level the model tends to think more per turn, so re-run your effort sweep instead of carrying a setting over.
  • Thinking cannot be turned off and the raw chain of thought is never returned. reasoning_effort is the only lever.
  • The model accepts max_tokens and effort only. The gateway strips sampling parameters such as temperature and top_p before forwarding, so a request that sets them still succeeds.
  • Anthropic rejects forced tool use — tool_choice: "required" or a named function — with a 400. The gateway forwards these requests with auto instead, so existing code keeps working; use response_format with a JSON schema when you need schema-valid output.
  • Thinking blocks are tied to the model that produced them. A conversation that moves from Claude Opus 5 to Claude Opus 5.5 keeps its reasoning; one that arrives from a Claude Fable model continues without it.
1curl https://api.llmgateway.io/v1/chat/completions \2    -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3    -H "Content-Type: application/json" \4    -d '{5        "model": "anthropic/claude-opus-5-5",6        "reasoning_effort": "high",7        "messages": [{ "role": "user", "content": "Review this pull request for concurrency bugs." }]8    }'

The model routes through Anthropic directly for now. Bedrock, Vertex AI, and Microsoft Foundry mappings follow once we have verified them end to end.

Provider cache control docs → | Claude Opus 5.5 on the models page →