Claude Sonnet 5.5 breaks Sonnet 5 code in five ways, and one fails silently
Anthropic lists five ways that code written for Claude Sonnet 5 can break on Claude Sonnet 5.5, and the one that reports nothing is the change to thinking blocks. Sonnet 5.5's thinking blocks work only in the account that produced them, and when another account sends one, Anthropic's API release notes say the API "drops the block before the model sees it, and the request succeeds".

A Claude Code session answering "can you add tests for this file?" by reading localizationUtils.ts and writing a new test file, with a "Pondering..." status underneath. Image: Anthropic Image: Anthropic
Nothing in the reply flags the missing block, and a call that used to carry Claude's reasoning forward comes back as an ordinary success built without it.
Anthropic ties the binding to distillation. Sonnet 5.5 "expands preserved thinking, so Claude's thinking cannot be decoupled from the account that created it", the company says, describing distillation attacks as attackers using thousands of fake accounts to extract a model's capabilities at industrial scale.
Blocks produced by earlier models are unaffected, and an account linked to the one that made the block still works. Anthropic's own line is that "most developers won't notice a change".
Claude Code sessions are the case Anthropic names
The company points at one situation in particular: moving conversations between accounts, including switching accounts mid-session in Claude Code.
Anthropic's migration guide sets a second rule for the same blocks. Each Sonnet 5.5 thinking block is signed over the conversation before it, and on accounts created on or after August 31, 2026 the API enforces that by default on the Claude API, Amazon Bedrock and Google Cloud, so a request that replays a block after an edit to earlier history returns a 400 error. The advice is to keep conversations append-only and to change instructions or tools with mid-conversation system messages.
Sonnet 5.5 arrived on September 28 at $2 per million input tokens and $10 per million output tokens, with a 1M-token context window and cache reads at $0.20 per million tokens. It sits under Claude Opus 5.5 at $4 and $20 per million tokens, and Claude Code has defaulted to Opus on the $20 Pro plan since September 22.
Four other changes return an error or go quiet
Anthropic's overview of Sonnet 5.5 lists the other four:
- Turning off up-front thinking means sending thinking: {"type": "between_tools"} instead of "disabled", at high effort or below.
- Forced tool use, the tool_choice types any and tool, returns a 400 error.
- On the Claude API and Google Cloud, the earlier computer_20251124 computer use tool is not accepted.
- The advisor tool rejects Claude Opus 4.8, Claude Opus 4.7 and Claude Sonnet 5 as advisors.
A sixth change alters the shape of a response without failing anything. Text between tool calls comes back inside thinking blocks, and Anthropic warns that an application streaming that text to its users "goes quiet between tool calls" until it sets a display value that returns the text, or turns off up-front thinking with between_tools.
Older code has two more errors to clear
Anyone moving from Claude Sonnet 4.6 or earlier hits two further 400s, according to Anthropic's migration guide. Thinking budgets are refused with the message that "thinking.type.enabled" is not supported for this model, and a non-default temperature, top_p or top_k value is refused as well.
The guide's advice on replacing a budget is to set an effort level and then test it: "There's no fixed mapping from a budget to an effort level, so run your evaluations at two or three levels." Anthropic puts Sonnet 5.5's retirement no sooner than September 28, 2027, which is how long there is to get the errors out.