Summary:

Anthropic is retiring both Claude Sonnet 4.5 API IDs on November 30, 2026. Check older n8n defaults and test the move to Sonnet 5.5.

Anthropic will retire Claude Sonnet 4.5 on the Claude API on November 30, 2026. The date applies to both claude-sonnet-4-5-20250929 and the shorter claude-sonnet-4-5. Anthropic recommends claude-sonnet-5-5. The retirement notice does not specify a cutoff hour. Check your provider separately if you use Amazon Bedrock or Google Vertex AI; those platforms set their own dates.

The model ID is the first thing to change, but not the only one. Older n8n nodes can pick Sonnet 4.5 by default, and Claude Sonnet 5.5 handles reasoning, tool calls and responses differently.

If you use the Anthropic Chat Model node in n8n

In the n8n 2.37.7 release source, version 1.3 of the Anthropic Chat Model node defaults to claude-sonnet-4-5-20250929. That is a node version, not the version of the n8n application. It also does not tell you which model an existing workflow actually selects: an explicit choice or expression can override the default. A maintainer recommended upgrading the node rather than expecting n8n to rewrite a released default.

  1. Find the actual model. Check the Model field, and any expressions or configuration it reads. Look for both retiring IDs in any fallback path too.
  2. Change a test copy first. Keep copied triggers and schedules inactive. Update the node if needed, then explicitly select Sonnet 5.5 or enter its ID where the node and credentials support it. A newer node doesn’t necessarily default to 5.5.
  3. Replay a few typical runs. Test a plain-text request and, where relevant, a tool call, streaming response or structured output. Check that the next node still gets the data it expects. Then apply the tested change to the active workflow.

See the n8n node reference for available model and node options.

If you call the Claude API directly

  • Search for both IDs. Check application settings, hard-coded scripts, jobs and fallback lists. Test claude-sonnet-5-5 on representative requests before switching production traffic.
  • Revisit thinking settings and tokens. Sonnet 5.5 uses adaptive thinking if you omit thinking. Remove old fixed budgets and disabled. If you want to skip up-front thinking, use between_tools at low, medium or high effort. With tools, thinking blocks can still appear. Recheck max_tokens, which covers thinking as well as the answer.
  • Read response blocks by type. Don’t assume the first block is text. When sending conversation history back during a tool loop, preserve the thinking blocks unchanged. Check streaming output as well as the final response.
  • Retest legacy request options. Remove custom temperature, top_p and top_k values, and replace assistant-prefill patterns. If you used forced tool selection, Anthropic recommends automatic tool selection with strict schemas. A strict schema controls the arguments if a tool is called; it doesn’t guarantee that the model calls it.

The official Sonnet 5.5 migration guide has separate notes for older beta headers, structured output and tool-specific integrations. Before you switch, check one real output shape, one representative tool or streaming path if you use it, and a fallback that does not point back to either retiring ID.