Error 400: Bad Request with GPT-o1 on Azure AI Foundry

Thank you for your answer @Prashanth Veeragoni !

I checked your propositions, but there is no menu item called connections, as shown here:

There is a menu item called "Connected resources" in the Management Center, although the edit option for these connections has no functionality to specify the API version, as shown in the following image:

This problem does not seem to be documented anywhere yet, could it be that it is actually a bug with the AI Foundry platform?

Christian Mittler,

Thanks for sharing the screenshots and detailed background — this issue is indeed subtle and not well documented yet. You're encountering this error:

Error 400: Bad Request ... 'Model {modelName} is enabled only for API versions 2024-12-01-preview and later'

Even after specifying the correct api_version in the flow.dag.yaml, it still fails when triggered via Prompt Flow in Azure AI Foundry, while working fine in the Playground. This hints that Prompt Flow is not yet properly passing or honoring the api_version setting, particularly for newer models like GPT-4o or GPT-o1.

Root Cause:

·       The api_version for Azure OpenAI is not automatically inferred from the model version.

·       The current Prompt Flow UI or YAML handling does not yet support injecting api_version at the required layer (the connection level).

·       In the “Connected Resources” (screenshot 2), there is no field for api_version, which confirms that the Azure AI Foundry's Prompt Flow implementation is hardcoding or defaulting the API version, likely to a version older than 2024-12-01-preview.

Solution:

You cannot currently fix this through Prompt Flow UI alone, but you can override it via Python code using the @tool decorator or via a custom component where you manually call the OpenAI chat API and pass the correct API version.

Use a Custom Python Tool to Call Azure OpenAI API

Here’s a code snippet that creates a callable node that explicitly sets api_version.

from promptflow import tool
from openai import AzureOpenAI
@tool
def chat_with_gpt_o1(question: str, chat_history: list) -> str:
    client = AzureOpenAI(
        api_key="YOUR_API_KEY",  # Use secrets in production
        api_version="2024-12-01-preview",
        azure_endpoint="https://<your-resource-name>.openai.azure.com/"
    )
    messages = [{"role": "system", "content": "You are a helpful assistant."}]
    for chat in chat_history:
        messages.append({"role": chat["role"], "content": chat["content"]})
    messages.append({"role": "user", "content": question})
    response = client.chat.completions.create(
        model="o1",  # or your deployment name
        messages=messages,
        temperature=0.7,
        max_tokens=256,
        top_p=1
    )
    return response.choices[0].message.content

Then use this tool in your Prompt Flow.

Hope this helps.

Thanks. 

Hi Christian Mittler,

Following up to see if the above suggestion was helpful. And, if you have any further query do let me know.

Thank you!

Hi Christian Mittler,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution, please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Thank you!

Thank you very much @Prashanth Veeragoni !

Your solution solved the problem, although this should only be temporary, until the issue is resolved.

Although another issue arrived, as I currently cannot deploy any Prompt Flows whatsoever, the platform is returning the following error:

The request is invalid. Status: 400 (The request is invalid.) ErrorCode: BadRequest Content: { "error": { "code": "BadRequest", "message": "The request is invalid.", "details": [ { "code": "InferencingClientCallFailed", "message": "{\"error\":{\"code\":\"Validation\",\"message\":\"{\\\"errors\\\":{\\\"\\\":[\\\"This endpoint has not been created successfully or is in deleting provisioning state. Please recreate endpoint and then try again to create deployment.\\\"]},\\\"type\\\":\\\"https://tools.ietf.org/html/rfc9110#section-15.5.1\\\",\\\"title\\\":\\\"One or more validation errors occurred.\\\",\\\"status\\\":400,\\\"traceId\\\":\\\"00-21faad9a571c440e86211a3746584f41-2ce498690ea2653f-01\\\"}\"}}", "details": [], "additionalInfo": [] } ], "additionalInfo": [ { "type": "ComponentName", "info": { "value": "managementfrontend" } }, { "type": "Correlation", "info": { "value": { "operation": "21faad9a571c440e86211a3746584f41", "request": "f41f8996739e347c" } } }

Hi Christian Mittler,

This new error clearly points to a platform-side issue with the Azure AI Endpoint — not the Prompt Flow itself.

The key part of the error is: "This endpoint has not been created successfully or is in deleting provisioning state. Please recreate endpoint and then try again to create deployment."

What This Means:

·       The endpoint backing your Prompt Flow is either:

o   Still provisioning,

o   Failed during creation,

o   Or is stuck in a deleting state and not yet cleaned up.

Prompt Flow is trying to deploy or call a managed endpoint, but the endpoint is not usable.

Solution Steps:

Step1: Go to Azure Portal → Azure Machine Learning → Endpoints

·       Check if the endpoint is:

o   Stuck (status = deleting, creating, or failed)

o   Or deleted but still listed

Step2: Try Deleting the Broken Endpoint

·       Manually delete the problematic endpoint.

·       If it is stuck in "deleting" and cannot be deleted manually:

o   Wait 15–20 minutes — Azure may take time to finish cleanup.

o   Alternatively, use Azure CLI to force the deletion:

az ml online-endpoint delete --name <your-endpoint-name> --resource-group <your-rg> --workspace-name <your-ws>

Step3: Recreate the Endpoint

·       Once it's fully deleted, redeploy your Prompt Flow or model.

·       Make sure the deployment target is healthy (compute instance, AKS, etc.)

Hope this helps.

Thank you!