how to use azureOpenAI GPT-5 (Responses API)

[nonakahiroshi-sys]

Hello,

I have been using Azure OpenAI models from a locally hosted Docker LibreChat instance. Mainly, I have been using the router model for GPT-4.

Since the GPT-5 model was released, I tried to use it, but it didn’t work properly.

What I did:

• I was using version librechat-0.7.8, and just like before, I added the gpt-5 model and tried chatting, but I got the following error:

Something went wrong. Here's the specific error message we encountered: An error occurred while processing the request: 400 Unsupported parameter: 'messages'. In the Responses API, this parameter has moved to 'input'. Try again with the new parameter. See the API documentation for more information: https://platform.openai.com/docs/api-reference/responses/create.

• From the error message, it seemed that Responses API support was required, so I updated to librechat-0.8.0-rc2.
• When using GPT-5 on 0.8.0-rc2 with the “Use Responses API” switch ON, the error message changed to:

Something went wrong. Here's the specific error message we encountered: An error occurred while processing the request: 404 Resource not found Troubleshooting URL: https://js.langchain.com/docs/troubleshooting/errors/MODEL_NOT_FOUND/

The URL and key are exactly as shown on Microsoft’s site, but to verify, I tested with curl and it worked:

$ curl https://<user-snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=2025-04-01-preview \
    -H "Content-Type: application/json" \
    -H "api-key: $AZURE_API_KEY" \
    -d '{
          "model": "gpt-5",
          "input": [
            {"role": "user", "content": "Hello!"}
          ]
        }'
{
  "id": "resp_68a6e7705a5c8191853df9a9ddccfc2e0df5ef7e8ce9bf87",
  "object": "response",
  "created_at": 1755768689,
    ...<snip>...

Is there something wrong with my configuration?

Here are my current settings:

  azureOpenAI:
    titleConvo: true # Enable title conversation for Azure OpenAI
    # Endpoint-level configuration
    plugins: true
    imageGeneration:
      enabled: true
      model: "dall-e-3"
      azure: true
      deploymentName: "dall-e-3"
    groups:
    # Group-level configuration
      - group: "ROUTER"
        instanceName: "<resource-snip>" # name of the resource group or instance
        baseURL: "https://<user-snip>-eastus2.cognitiveservices.azure.com/openai/deployments/model-router/chat/completions?api-version=2025-01-01-preview"
        apiKey: "<KEY>"
        models:
          model-router:
            deploymentName: model-router
            version: "2025-01-01-preview"
      - group: "CBU"
        instanceName: "<resource-snip>" # name of the resource group or instance
        baseURL: "https://<user-snip>-swedencentral.cognitiveservices.azure.com/openai/deployments/gpt-4.1/chat/completions?api-version=2025-01-01-preview"
        apiKey: "<KEY>"
        models:
          gpt-4.1:
            deploymentName: gpt-4.1
            version: "2025-01-01-preview"
      - group: "CBU2"
        instanceName: "<resource-snip>" # name of the resource group or instance
        baseURL: "https://<user-snip>-eastus2.cognitiveservices.azure.com/openai/deployments/o4-mini/chat/completions?api-version=2025-01-01-preview"
        apiKey: "<KEY>"
        models:
          o4-mini:
            deploymentName: o4-mini
            version: "2025-01-01-preview"
      - group: "GPT5"
        instanceName: "<resource-snip>" # name of the resource group or instance
        baseURL: "https://<user-snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=2025-04-01-preview"
        apiKey: "<KEY>"
        models:
          gpt-5:
            deploymentName: gpt-5
            version: "2025-04-01-preview"

Any help would be greatly appreciated.
Thank you!

[nonakahiroshi-sys]

I tried overriding global.fetch to check the logs, and it seems that the request destination is not set as expected.

The URL that works when sending via curl (and set in librechat.yml):

https://<user-snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=2025-04-01-preview

The URL accessed during chat operation:

https://<user-snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=preview

The api-version is incorrect... why?

[nonakahiroshi-sys]

I'm not very familiar with this kind of code, so I can't quite tell where the URL is being modified…

As a brute-force workaround, I just wanted to report that I was able to get it working by overriding global.fetch as shown below:

const originalFetch = global.fetch;
global.fetch = async (input, init) => {
  let url = input instanceof Request ? input.url : input;
  let bodyPreview = null;

  try {
    if (input instanceof Request) {
      const clone = input.clone();
      bodyPreview = await clone.text();
    } else if (init?.body) {
      if (typeof init.body === "string") {
        bodyPreview = init.body;
      } else if (init.body instanceof Buffer || init.body instanceof Uint8Array) {
        bodyPreview = Buffer.from(init.body).toString("utf8");
      } else {
        bodyPreview = `[unhandled body type: ${init.body.constructor?.name}]`;
      }
    } else {
      bodyPreview = "[no body]";
    }
  } catch (err) {
    bodyPreview = `[Error reading body: ${err.message}]`;
  }

  console.log("=== FETCH CALLED ===");
  console.log("URL:", url);
  console.log("Body:", bodyPreview);
  console.log("====================");

  if ( url === "https://<snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=preview" )
  {
    console.log("MOD URL!");
    if ( input instanceof Request )
    {
      input.url = "https://<snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=2025-04-01-preview";
    }
    else
    {
      input = "https://<snip>-eastus2.cognitiveservices.azure.com/openai/responses?api-version=2025-04-01-preview";
    }
  }

  return originalFetch(input, init);
};

[kinzayed]

Hey,

I'm facing the same issue after adding GPT-5, can you please show where you added your global.fetch code?

Thanks!

[kinzayed]

I actually managed to fix this error by changing the librechat.yaml config:

- group: "gpt5"
  apiKey: "${GPT-5-API-KEY}"
  baseURL: "https://${INSTANCE_NAME}.openai.azure.com/openai/v1"
  instanceName: "${INSTANCE_NAME}"
  version: "preview"
  models:
    gpt-5:
      deploymentName: "gpt-5"
      useOverrides: true

Also, under modelSpecs, make sure to use

preset:
  useResponsesApi: true

[nonakahiroshi-sys]

I see?

Now, it works without a patch by specifying the resource endpoint rather than the deployment endpoint!

It looks like this:

RESOURCE_NAME is set as ${INSTANCE_NAME}-eastus2.

  - group: "eastus2"
    instanceName: "${INSTANCE_NAME}" # name of the resource group or instance
    baseURL: "https://${RESOURCE_NAME}.openai.azure.com/openai/v1"
    apiKey: "${GPT5-API-KEY}"
    models:
      gpt-5:
        deploymentName: gpt-5
        version: "2025-04-01-preview"

So, in this case, the problem is that “it does not work when you specify the deployment endpoint”.