Azure OpenAI Response API request fails

[normunds-wipo]

What happened?

According to this https://github.com/danny-avila/LibreChat/pull/8195 it should be implemented, but we do not manage to use it. Here is our setup:

endpoints:
  azureOpenAI:
    groups:
      - group: response
        apiKey: "${AZURE_FOUNDRY_API_KEY}"
        instanceName: "oai-ai-assistant-response-sc"
        baseURL: https://oai-ai-assistant-dev-sc.openai.azure.com/openai/responses
        version: 2025-04-01-preview
        models:
          o3-pro:
            deploymentName: "o3-pro"

 modelSpecs:
  list:
    - name: o3-pro
          label: o3-pro (azure, response)
          preset:
            modelLabel: AI Assistant (azure)
            endpoint: azureOpenAI
            model: o3-pro
            useResponsesApi: true

There are two problems. First is that version value is not used and always replaced with "preview". Why?
in packages/api/src/endpoints/openai/llm.ts there is code:

      configOptions.defaultQuery = {
        ...configOptions.defaultQuery,
        'api-version': configOptions.defaultQuery?.['api-version'] ?? 'preview',
      };

it seems that configOptions.defaultQuery does not get created correctly and is always an empty object (so we always get "preview" as the version). Now if I patch this, so that it uses the correct API version, the endpoint seems to be computed correctly as it gets an answer from API. But it always is the same error:

There is nothing wrong with the model endpoint or credentials, we can hit it succesfully from standalone code. Though we might be using different payload - a bit hard to compare.

Version Information

v0.8.0-rc3

Steps to Reproduce

• set up the model configuration as above
• patch problem 1 (else it just fails to connect)
• run LibreChat - we run from source code
• any request to model ends with a 400 error - "Invalid 'response_id': 'responses'. Expected an ID that contains letters, numbers, underscores, or dashes, but this value contained additional characters."

What browsers are you seeing the problem on?

No response

Relevant log output

2025-09-11 10:48:02 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted 400 Invalid 'response_id': 'responses'. Expected an ID that contains letters, numbers, underscores, or dashes, but this value contained additional characters.

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

There are two problems. First is that version value is not used and always replaced with "preview". Why?

Azure explains this here: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/api-version-lifecycle?tabs=key#api-evolution

---

The following config works for me:

# librechat.yaml

azureOpenAI:
# Group level config under `azureOpenAI`
    - group: "arbitrary-group-name"
      apiKey: "${YOUR_API_KEY}" # .env var reference
      instanceName: "your-instance-name"
      version: "2025-04-01-preview" # whichever version works for non-responses
      models:
        o3-pro:
          deploymentName: "o3-pro"

# librechat.yaml

modelSpecs:
  list:
  - name: o3-pro
    label: o3-pro (azure, response)
    preset:
      modelLabel: AI Assistant (azure)
      endpoint: azureOpenAI
      model: o3-pro
      useResponsesApi: true

---

You need to omit baseURL

# librechat.yaml

endpoints:
  azureOpenAI:
    groups:
      - group: response
        apiKey: "${AZURE_FOUNDRY_API_KEY}"
        instanceName: "oai-ai-assistant-response-sc"
# -- baseURL: https://oai-ai-assistant-dev-sc.openai.azure.com/openai/responses <--------- omit this 
        version: 2025-04-01-preview
        models:
          o3-pro:
            deploymentName: "o3-pro"

[normunds-wipo]

Thanks a lot. So you assume that people will use v1 API, and that's why you on purpose replace any api-version with "preview"?
IMO it should work also with:
baseURL: https://oai-ai-assistant-dev-sc.openai.azure.com/openai
version: 2025-04-01-preview
as-is it does not (because "2025-04-01-preview" gets replaced with "preview" in lmm.ts as per my post)
But otherwise I managed to get it working!

[danny-avila]

Glad you got it working. The v1 API is required for this to work:
https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/responses?tabs=python-key