openai[patch]: allow specification of output format for Responses API

[ccurme]

Add an output_version attribute to BaseChatOpenAI. Motivation is to allow users to opt-in to breaking changes in AIMessage formats.

"v0" is the default and corresponds to current format. We intend to introduce standard types for reasoning, citations, and other AIMessage content with "v1". At that point we will add the attribute to BaseChatModel.

Here we implement "responses/v1" to allow users to opt-in to the change described in #31587 — this is a breaking change that is necessary to support some features (e.g., remote MCP tool use under zero data retention contexts).

[vercel]

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

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Jun 20, 2025 5:47pm

[codspeed-hq]

CodSpeed WallTime Performance Report

Merging #31686 will not alter performance

_{Comparing cc/output_version (120d891) with master (7ff4050)}

⚠️ 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

✅ 13 untouched benchmarks

[eyurtsev]

Can we remove this until there's more than one version?

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #31686 will not alter performance

_{Comparing cc/output_version (120d891) with master (7ff4050)}

Summary

✅ 13 untouched benchmarks

[sydney-runkle]

This makes a lot of sense - thanks for the clean docs, etc.

A few follow up questions:

• Can we make the default v1 when we bump version to v1?
• I presume v1 will be using a new form of standard output?

A higher level question: isn't our job to support a uniform output format across all models? In that sense, shouldn't we not support responses format specifically? That seems quite tied to openai... perhaps this just necessitates the shift towards, for lack of a better word, stdout 😉