claude-sonnet-5-5 — pricing, context window and limits

Same price as Sonnet 5, and it caches shorter prompts

Sonnet 5.5 is the current Sonnet. It kept Sonnet 5's per-token prices and tokenizer, so a move between them costs nothing on the rate card and needs no token recount. What it changed on the cost side is caching: its minimum cacheable prefix is lower than Sonnet 5's, so system prompts that were too short to cache before can start producing cache reads with no code change. Check the usage block after switching — it is one of the few migrations where the bill can fall on its own.

The off switch moved

thinking: {type: "disabled"} returns a 400 here. The lowest setting is now between_tools, which does no extended thinking and returns the model's short progress notes between tool calls. It comes with limits of its own: it is accepted only at effort high or below, takes no other field inside thinking, cannot be combined with a per-message effort change, and exists on this model only — any other model rejects it. Anthropic's advice is to try adaptive thinking at low effort before reaching for it. See disabled thinking rejected on Sonnet 5.5.

Effort levels were recalibrated

The default is still high, but a level no longer produces the same amount of thinking it did on Sonnet 5. The documented starting points are medium for agentic coding and multistep tool use, and low for chat, extraction and classification. Re-run your effort sweep instead of reusing the Sonnet 5 setting — the level names survived the change, the meaning didn't.

The rest of the breaking changes

Forced tool choice returns a 400. Thinking blocks are tied to this model and conversation — no other model reads them. Computer use on the Claude API and Google Cloud needs the computer toolset. And the advisor tool accepts fewer advisor models. Declines arrive in more categories than on Sonnet 5, and server-side fallback retries only some of them.

Related

See migrating: Sonnet 5 to Sonnet 5.5 and Sonnet 5.5 vs Haiku 4.5.