Claude Sonnet 5.5: what the September release means

Anthropic announced Claude Sonnet 5.5 on September 28, 2026, positioning it as a faster, lower-cost complement to Opus 5.5 for everyday work. The API model name is claude-sonnet-5-5. Haiku 5.5 remains a planned release for the coming weeks, not a launched product in this announcement. Official release.

The practical interest is frequent iteration: fixing a contained bug, revising a document or preparing a spreadsheet. Vendor performance claims do not establish how much faster your particular job will finish, because checking and correcting the output remain part of that job.

An API change worth checking before switching

If your integration used thinking: {"type": "disabled"}, changing only the model name is insufficient. Sonnet 5.5 rejects that setting. Its migration guide specifies between_tools for disabling up-front thinking, at high effort or below; higher effort requires adaptive thinking. Migration instructions.

This matters most to developers maintaining existing integrations. An invalid-request error immediately after a model swap may be a configuration mismatch rather than a broken user prompt. Review the request settings and the migration guide before rewriting the application's instructions. The guide covers other compatibility changes too; the setting above is not a complete migration recipe.

Claude Code users should keep the application and model names separate. Claude Code is the coding tool, while Sonnet 5.5 is a model it can use. An application's version, selected model and repository state all matter when reproducing a result. Claude Code overview.

A useful comparison takes one familiar task

For an illustrative test, choose a signup form whose button remains disabled after a failed request. Supply the same failing case to the previous and new setup. Success means the visitor can retry after an error, repeated clicks during submission do not duplicate the request and the existing appearance remains intact.

That test can reveal whether faster iteration helps: how many corrections were needed, whether the change stayed within scope, and whether the final browser behavior matched the request. It also exposes a bad shortcut, such as leaving the button enabled throughout submission. A large diff or confident final message would not settle the result.

The purchase decision can wait for the evidence

Try the new model on work with a clear finish before changing every default or adding another subscription. Check availability and billing in your own account; this report does not verify every regional plan. No independent model benchmark was performed here.

Read the Claude Code workflow guide for a repeatable evaluation, and the Opus release coverage for the more demanding-work option. This article was published September 29, one day after the Sonnet announcement.