Investigation: Assess OpenAI Responses API migration

[mattprintz]

Context

OpenAIModel constructs langchain_openai.ChatOpenAI. For ordinary models it currently relies on LangChain routing, which uses the Chat Completions API unless Responses-only options are present. Archytas supplies reasoning for supported reasoning models, so those requests route to the Responses API; ordinary models such as the default gpt-4o continue to use Chat Completions.

OpenAI recommends the Responses API for new work. The Chat Completions API remains supported; this issue should verify OpenAI's current policy and any applicable retirement timeline rather than assume a deprecation date.

Scope

Assess and, if appropriate, implement a deliberate Responses API path for native OpenAI models.

The assessment must account for the different compatibility requirements of:

• native OpenAI (OpenAIModel);
• OpenAI-compatible endpoints configured with base_url;
• OpenRouter, which documents Chat Completions compatibility; and
• Azure OpenAI.

Acceptance criteria

• Document the applicable OpenAI lifecycle guidance and whether a migration is required.
• If a change is appropriate, native OpenAI requests use the intended endpoint by explicit configuration rather than implicit LangChain routing.
• Agent tool-call loops, multi-turn history, text extraction, and reasoning summaries work for both ordinary and reasoning OpenAI models.
• OpenAI-compatible providers do not regress because they lack Responses API support.
• Add automated coverage for the selected routing and response shapes, and update user-facing configuration documentation if behavior changes.