ChatOpenAI integration: parameters and OpenAI-compatible endpoint setup

TL;DR: The OpenAI integration ships in langchain-openai. Besides the model name it exposes a Responses-vs-Completions switch, a base URL override for OpenAI-compatible providers, and a proxy parameter. Never paste a real key into code or chat. - For streaming token usage metadata from OpenAI, you must opt in on the integration; usage is not returned in streaming chunks by default.

The fix

1. use_responses_api - picks the OpenAI Responses API vs the Chat Completions API.
2. base_url - point at an OpenAI-compatible endpoint (Together, vLLM, and similar) while keeping model_provider="openai" semantics. Provider-specific fields from routers may not be preserved through this path.
3. openai_proxy - HTTP proxy for deployments that need one.
4. max_retries / timeout - resilience (see the retry skill). Defaults: up to 6 retries on 429/5xx/network errors.
5. OpenRouter: ChatOpenRouter from langchain-openrouter
6. LiteLLM: ChatLiteLLM / ChatLiteLLMRouter from langchain-litellm Do not point ChatOpenAI at a router's base URL and expect router-specific fields to survive. The docs call this out explicitly.
7. The key comes from OPENAIAPIKEY in the environment. Never paste a real key into code or chat.
8. For streaming token usage metadata from OpenAI, you must opt in on the integration; usage is not returned in streaming chunks by default.

When to use this

• This covers exactly what the title says: ChatOpenAI integration.
• You are setting this up for the first time, or auditing an existing setup.
• You want the key gotchas in one place before you start.

When not to use this

• You are doing a different workflow with ChatOpenAI; these steps are specific to the title above.
• You need the full reference docs; this is the short path, not the manual.

Compatibility

• Mentions 5.5 in the original; behavior may differ on much older or newer releases.

Verified against the official LangChain docs on 2026-09-26. Models: https://docs.langchain.com/oss/python/langchain/models