fix(openai): avoid PydanticSerializationUnexpectedValue for structured output

[Jason Meng (JasonCZMeng)]

Summary

Exclude the parsed field from model_dump() in _create_chat_result() to prevent PydanticSerializationUnexpectedValue warnings when using structured output.

Fixes #35538

Root Cause

When OpenAI returns a parsed completion, message.parsed holds an arbitrary Pydantic BaseModel instance (the user's structured output schema). The existing code calls response.model_dump() which tries to serialize this into a dict — but the OpenAI SDK's ParsedChatCompletion types the parsed field as None, so Pydantic emits a serialization warning when it encounters a real model there.

Fix

Exclude parsed from the model_dump() call since it's already copied separately into additional_kwargs["parsed"] from the typed response object (at ~L1602). No data is lost.

Changes

libs/partners/openai/langchain_openai/chat_models/base.py

• Added exclude={"choices": {"__all__": {"message": {"parsed"}}}} to the model_dump() call in _create_chat_result()

libs/partners/openai/tests/unit_tests/chat_models/test_base.py

• Added test_create_chat_result_avoids_parsed_model_dump_warning() — mocks an OpenAI completion with a BaseModel in parsed and asserts no serialization warning is emitted

[codspeed-hq]

Merging this PR will not alter performance

✅ 6 untouched benchmarks
⏩ 30 skipped benchmarks¹

---

_{Comparing JasonCZMeng:fix/issue-35538 (9bbea53) with master (e50625e)}

Footnotes

1. 30 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩