[kimnamu]

Thanks for merging #7411 so quickly, @longcw. Claude Sonnet 5.5 rejects a forced toolChoice on Bedrock the same way Opus 5.5 does, but _MODELS_REJECTING_FORCED_TOOL_CHOICE only lists claude-opus-5-5 and claude-fable-5-1. So tool_choice="required" or a named tool still goes out as any/tool, and Bedrock returns a 400. This adds claude-sonnet-5-5 to that tuple. The sampling-params gate already covers Sonnet 5.5 through its claude-sonnet-5 substring.

case	Before	After
us.anthropic.claude-sonnet-5-5, tool_choice="required"	❌ ValidationException (400)	✅ sent as auto, the model called get_weather
same, named tool get_weather	❌ ValidationException (400)	✅ sent as auto, the model called get_weather
Opus 5.5, Fable 5.1	✅ auto	✅ unchanged
Opus 5, Sonnet 5, Fable 5, Sonnet 4.6: required → any, named → tool	✅	✅ unchanged (Sonnet 5 row added to the test)

Problem

ValidationException: An error occurred (ValidationException) when calling the ConverseStream operation:
The model returned the following errors: tool_choice: type "tool" and "any" are not supported for this model.

Fix

One entry in _MODELS_REJECTING_FORCED_TOOL_CHOICE. The substring match doesn't reach claude-sonnet-5, which still accepts any (now covered by test_forced_tool_choice_kept_for_other_models).

Tests

$ uv run pytest tests/test_plugin_aws_llm.py -q
16 passed in 1.50s
$ make type-check
✓ Type checking passed
$ uv run ruff check livekit-plugins/livekit-plugins-aws tests/test_plugin_aws_llm.py
All checks passed!

RED on main (source reverted, new test rows kept) and live runs

$ uv run pytest tests/test_plugin_aws_llm.py -q      # llm.py from main
>           assert await _tool_choice(model, "required") == {"auto": {}}
E           AssertionError: assert {'any': {}} == {'auto': {}}
1 failed, 15 passed in 0.47s

Live, us-east-1, aws.LLM(model="us.anthropic.claude-sonnet-5-5").chat(..., tools=[get_weather], tool_choice=...):

main     tool_choice='required'          sent={'any': {}}                   -> APIStatusError: ... tool_choice: type "tool" and "any" are not supported for this model. status_code=400
main     tool_choice=<named get_weather>  sent={'tool': {'name': 'get_weather'}} -> APIStatusError: ... status_code=400
this PR  tool_choice='required'          sent={'auto': {}}                  -> OK tool_calls=['get_weather']
this PR  tool_choice=<named get_weather>  sent={'auto': {}}                  -> OK tool_calls=['get_weather']

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[kimnamu]

Thanks @longcw for reviewing #7411, which added the forced tool_choice list that this PR extends with Claude Sonnet 5.5.

State	open, no conflicts with main
CI	all checks green on head 09274a6
main @ 296f28b	Sonnet 5.5 with tool_choice="required" still gets a 400 ValidationException (live, us-east-1). With this branch applied, tests/test_plugin_aws_llm.py: 16 passed

Would you mind taking a look when you have a moment? If this isn't a priority right now, tell me and I'll close it to keep your queue clean.