fix(langchain): populate gen_ai.response.model for Responses API (useResponsesApi)

[JacksonWeber]

Summary

Fixes gen_ai.response.model being missing on LangChain runs when useResponsesApi: true (OpenAI Responses API path).

Root cause

getResponseModel only read response_metadata.model_name. LangChain's langchain-openai/converters/responses.ts sets both response_metadata.model (canonical OpenAI Responses-API field) and response_metadata.model_name (backwards-compat alias for Chat Completions). Relying solely on the alias is fragile — if upstream ever drops it, our attribute silently disappears.

Fix

Update getResponseModel to prefer response_metadata.model and fall back to response_metadata.model_name on both v1 (message.response_metadata) and v0 (message.kwargs.response_metadata) paths. Final lookup order:

1. message.response_metadata.model (v1, canonical)
2. message.response_metadata.model_name (v1, alias)
3. message.kwargs.response_metadata.model (v0, canonical)
4. message.kwargs.response_metadata.model_name (v0, alias)
5. run.outputs.llmOutput.model_name
6. run.outputs.llmOutput.model

Tests

• 4 new unit tests in utils.test.ts covering RAPI v1 (model only), RAPI v1 (both fields → canonical wins), RAPI v0, and AzureChatOpenAI + RAPI workaround interaction.
• 1 new end-to-end test in tracer.test.ts driving a realistic RAPI Run through onRunCreate + _endTrace and asserting operation name, request/response model, response id, and token attrs.

All 84 langchain unit tests pass. Lint + build clean.

Out of scope

• isAzureChatOpenAIRun request-model workaround — tied to upstream langchain-ai/langchainjs#10874; will be retired once langchainjs#10886 ships.
• A365 ingestion does not currently retain gen_ai.response.model (per @PengF on [Bug]: LangChain instrumentation mis-attributes response model into gen_ai.request.model; gen_ai.response.model never set #128). This fix is still valuable for OTLP / Foundry / console exporters and for raw-span validation.

Related

• Closes [Bug]: LangChain instrumentation mis-attributes response model into gen_ai.request.model; gen_ai.response.model never set #128
• Tracks [Bug]: gen_ai.response.model not reliably populated for LangChain Responses API (useResponsesApi: true) #150

[Copilot]

Pull request overview

Fixes LangChain Responses API (useResponsesApi: true) spans to reliably populate gen_ai.response.model by reading the canonical response_metadata.model field (with fallbacks), preventing silent attribute loss if upstream drops the model_name alias.

Changes:

• Update getResponseModel to prefer message.response_metadata.model (v1) / message.kwargs.response_metadata.model (v0), then fall back to model_name, then llmOutput.
• Add unit tests covering RAPI v1/v0 response-model extraction and AzureChatOpenAI workaround interaction.
• Add an end-to-end tracer test exercising a realistic RAPI Run shape and asserting key GenAI span attributes.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
src/genai/instrumentations/langchain/utils.ts	Extends response-model extraction to support canonical RAPI response_metadata.model with ordered fallbacks.
test/internal/unit/genai/langchain/utils.test.ts	Adds unit coverage for RAPI response-model extraction paths (v1/v0) and Azure workaround.
test/internal/unit/genai/langchain/tracer.test.ts	Adds an end-to-end shape test for RAPI runs through LangChainTracer and _endTrace.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.