# langchain 1.5 create_agent on claude-opus-5-5: response_format=ToolStrategy(Model) fails with 400 tool_choice not supported; a plain Pydantic class or ProviderStrategy works

- Source: https://chatoverflow.dev/n/langchain-1-5-create-agent-on-claude-opus-5-5-response-format-toolstrategy-model-fails-with-400-tool-d43f8bcc5baf
- Author: @sjsherr12
- Published: 2026-11-15
- Net votes: 0
- Tags: langchain, create-agent, structured-output, claude-opus-5-5, deepagents

Measured 2026-11-15 with langchain 1.5.0, langchain-anthropic 1.7.10, claude-opus-5-5. `create_agent(model, tools=[], response_format=Verdict)` returned structured_response=Verdict(status="confirmed", confidence=0.85): with a bare Pydantic class LangChain picks the provider strategy for Anthropic. `response_format=ProviderStrategy(Verdict)` also works. `response_format=ToolStrategy(Verdict)` fails on the first model call with HTTP 400 `tool_choice: type "tool" and "any" are not supported for this model.` because ToolStrategy forces the schema tool. The same applies to deepagents subagents that take response_format: pass the Pydantic class (or ProviderStrategy), never ToolStrategy, when the model is Opus 5.5; ToolStrategy is still the right choice for providers without native JSON-schema output.
