[Lucas Kim (kimnamu)]

Why are these changes needed?

Thanks for keeping create_stream working across so many OpenAI-compatible servers; the usage-chunk handling from #3972 is what this builds on.

create_stream reads finish_reason only from chunks with usage is None (_openai_client.py L948). The Amazon Bedrock runtime endpoint /openai/v1/chat/completions attaches usage to the GPT-6 finish chunk, so the streamed CreateResult.finish_reason is always "unknown", even for length and tool calls, while create() on the same request returns the real value.

This drops the chunk.usage is None condition. The stop_reason is None check that stays already keeps an earlier finish reason when a LiteLLM-style usage chunk follows. The Azure AI client already reads finish_reason whenever it is set.

	Before	After
Bedrock runtime us.openai.gpt-6-sol / us.openai.gpt-6-luna (and global.) stream: text, truncated, tool call¹	❌ unknown ×3 each	✅ stop / length / function_calls (same as create())
Bedrock runtime us.openai.gpt-6-astra / global.openai.gpt-6-astra stream: text, truncated	❌ unknown ×2 each	✅ stop / length (same as create())
openai.gpt-oss-120b-1:0, Mantle openai.gpt-6-sol / openai.gpt-6-astra (usage not on finish chunk)	✅	✅ unchanged
Streamed usage (GPT-6 Sol text)	(8, 5)	✅ (8, 5) unchanged
LiteLLM-style usage chunk after stop (existing tests)	✅	✅ unchanged
Public API, arguments, chunk order	—	✅ unchanged

¹ with reasoning_effort="none", which Bedrock chat completions needs for Sol/Luna function tools.

Test RED/GREEN, checks, live runs

New test test_openai_chat_completion_client_create_stream_finish_reason_with_usage (mocked: content chunk → finish chunk with usage → choices=[] usage chunk, stop and length):

# source reverted to main
E       AssertionError: assert 'unknown' == 'stop'
E       AssertionError: assert 'unknown' == 'length'
2 failed, 86 deselected in 2.68s
# with fix
2 passed, 86 deselected in 2.41s

tests/models/test_openai_model_client.py: main 2 failed / 53 passed (the new test), fix 55 passed, 33 skipped. The agentchat tests that use this client (test_assistant_agent.py, test_group_chat.py, test_declarative_components.py): 177 passed, 3 skipped. ruff format --check and ruff check clean, pyright 0 errors, mypy no issues on both files.

Live, base_url=https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1, max_completion_tokens=5:

# main
us.openai.gpt-6-sol: create=length create_stream=unknown
us.openai.gpt-6-luna: create=length create_stream=unknown
us.openai.gpt-6-astra: create=length create_stream=unknown
# this branch
us.openai.gpt-6-sol: create=length create_stream=length
us.openai.gpt-6-luna: create=length create_stream=length
us.openai.gpt-6-astra: create=length create_stream=length

Astra has no tool-call row: Bedrock's chat completions endpoint returns 400 for function tools with Astra ("To use function tools, use /v1/responses"), before and after this change.

Related issue number

Fixes #8277

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. See https://github.com/microsoft/autogen/blob/main/CONTRIBUTING.md to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.