!fix(core): _should_stream respects streaming param

[mdrxy]

Important

This is a breaking change

_should_stream() in BaseChatModel wasn't checking the instance-level streaming parameter, leading to streaming being enabled even when explicitly disabled

• Before: ChatOpenAI().stream() always attempted streaming (ignored streaming=False)
• After: ChatOpenAI().stream() respects streaming=False and falls back to invoke()

Any existing code doing:

model = ChatOpenAI()  # streaming=False by default
for chunk in model.stream(messages):  # This now calls invoke() instead of streaming!
    process(chunk)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Preview	Comments	Updated (UTC)
langchain	Ignored	Preview		Sep 8, 2025 8:21pm

[codspeed-hq]

CodSpeed WallTime Performance Report

Merging #32707 will improve performances by 13.02%

_{Comparing mdrxy/fix-oai-stream (8f1f9e4) with master (97dd762)}

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Summary

⚡ 5 improvements
✅ 8 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	test_import_time[BaseChatModel]	586.7 ms	519.1 ms	+13.02%
⚡	test_import_time[ChatPromptTemplate]	663.3 ms	602.7 ms	+10.06%
⚡	test_import_time[Document]	229.3 ms	206.9 ms	+10.86%
⚡	test_import_time[InMemoryRateLimiter]	182 ms	164.6 ms	+10.54%
⚡	test_import_time[PydanticOutputParser]	597.8 ms	542.4 ms	+10.2%

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #32707 will not alter performance

_{Comparing mdrxy/fix-oai-stream (8f1f9e4) with master (97dd762)}

Summary

✅ 14 untouched benchmarks

[ccurme]

why add these?

[mdrxy]

The unit tests needed updating because they were relying on the old behavior where the streaming parameter was ignored. Now they need to set streaming=True to test streaming. This is breaking - I'll update description to reflect.

[ccurme]

What is motivating this change?

I think the underlying issue is that streaming: bool = False on BaseChatOpenAI has no sentinel value— ideally it should be streaming: bool | None = None.

I think intended usage is that

llm = ChatOpenAI(model="...")
llm.stream(...)

streams without you having to explicitly specify streaming=True in the constructor. Otherwise this is (very) breaking and it's unclear what the benefit is.

[ccurme]

Can re-open if needed.