Azure openai in region sweden central seems not to be working.

Question

Error:

Traceback (most recent call last):

  File "/...../AI_Foundry/test_ai_foundry_gpt54.py", line 17, in <module>

    response = client.chat.completions.create(

               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/...../.pyenv/versions/3.12.11/lib/python3.12/site-packages/openai/_utils/_utils.py", line 287, in wrapper

    return func(*args, **kwargs)

           ^^^^^^^^^^^^^^^^^^^^^

  File "/...../.pyenv/versions/3.12.11/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py", line 925, in create

    return self._post(

           ^^^^^^^^^^^

  File "/...../.pyenv/versions/3.12.11/lib/python3.12/site-packages/openai/_base_client.py", line 1249, in post

    return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))

                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "/...../.pyenv/versions/3.12.11/lib/python3.12/site-packages/openai/_base_client.py", line 1037, in request

    raise self._make_status_error_from_response(err.response) from None

openai.BadRequestError: Error code: 400 - {'error': {'message': 'The requested operation is unsupported.'}}

---

AI Foundry provided test code for gpt-5.4-pro:

from openai import OpenAI

endpoint = "https://xxxxxxxxxxxxx-resource.openai.azure.com/openai/v1/"
deployment_name = "gpt-5.4-pro"
api_key = "xxxxxxxxxxx"

client = OpenAI(
    base_url=endpoint,
    api_key=api_key
)

completion = client.chat.completions.create(
    model=deployment_name,
    messages=[
        {
            "role": "user",
            "content": "What is the capital of France?",
        }
    ],
)

print(completion.choices[0].message)

Answer 1

Hello Truong, David,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

The observed behavior aligns with an API operation mismatch rather than a service outage or regional failure.After carefully reviewing the behavior, the response returned (400 – The requested operation is unsupported) confirms that the service endpoint is reachable, while the specific operation being invoked is not supported for the current request shape or API surface. This typically indicates a request‑level incompatibility rather than a regional or platform availability issue.

The deployment in Sweden Central is responding correctly at the network and service layer. A clean HTTP 400 response means the request is being actively validated and rejected, which differs from symptoms normally associated with regional outages such as timeouts or 5xx errors. This helps narrow the scope to API usage rather than availability.

Based on the current configuration, the most likely cause is a mismatch between the model deployment and the operation being called. As foundry supports multiple inference surfaces and given that not every model deployment behaves identically across all API operations, newer Foundry‑hosted models may expose inference primarily through the Responses API, which unifies chat‑style interactions with newer features.

To resolve the issue, please check if the following steps help

1. Please confirm the API surface being used
   • Ensure the base URL follows the OpenAI‑compatible v1 pattern: https://<resource-name>.openai.azure.com/openai/v1
   • When using the v1 API, dated api-version query parameters are not required.
   • Validate that the OpenAI SDK is being used against the v1 endpoint rather than a project or agent endpoint.
   • Please refer the following Azure OpenAI in Microsoft Foundry Models v1 API - Microsoft Foundry | Microsoft Learn Get started with Microsoft Foundry SDKs and Endpoints - Microsoft Foundry | Microsoft Learn
2. Validating operation compatibility
   • If the current call uses chat.completions, test the same deployment using the Responses API instead.
   • The Responses API is the recommended interface for newer Foundry deployments and supports chat‑style prompts using a unified request model.
   • Please refer the following - Use the Azure OpenAI Responses API - Microsoft Foundry | Microsoft Learn
3. Confirming deployment configuration
   • Please verify that the deployment name passed to the SDK exactly matches the configured deployment in the Foundry resource.
   • Ensure the deployment is in a ready state and not paused or quota‑restricted.
4. Ruling out regional constraints pragmatically
   • Model availability can vary by region and tenant, but an active deployment returning HTTP 400 indicates the region is functioning.
   • If verification is needed, deploying the same model to a second supported region and testing the identical request can help compare behavior.
   • Please refer the following Azure OpenAI in Microsoft Foundry Models REST API reference - Microsoft Foundry | Microsoft Learn Work with chat completion models - Microsoft Foundry | Microsoft Learn

Testing the deployment with the Responses API and validating the endpoint and deployment configuration should help moving forward.

Thank you

Answer 2

The error openai.BadRequestError: ... 'The requested operation is unsupported.' with status code 400 indicates that the request shape or operation is not supported for the target Azure OpenAI endpoint/deployment, rather than a pure network reachability issue from Sweden Central.

From the provided code and the available information, the following points are relevant:

1. Error meaning and class
   • HTTP 400 maps to BadRequestError in the Python client.
   • This is raised when the service receives the request but rejects it as invalid or unsupported.
   • It is not a 401/403 (auth/permission) or 404 (wrong path/model) or a connectivity error.
2. General troubleshooting guidance for 4xx errors
   • For 401/403: verify the identity or API key and that it has access to the Azure OpenAI resource.
   • For 404: verify that the endpoint uses the ...openai.azure.com/openai/v1/ path and that the model value is a valid deployment name.
   • For unexpected failures: check for proxy/firewall issues and try a smaller prompt.
   In this case, the endpoint format in the code matches the documented pattern:

      https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/
   

   and the call is using chat.completions.create, which is the correct operation type for chat models.
3. What to verify in this scenario Based on the documented troubleshooting steps, the following should be checked for this specific deployment:
   1. Endpoint and path
   • Confirm that the endpoint string exactly matches the Azure OpenAI resource endpoint from the Azure portal, including the /openai/v1/ suffix.
   2. Model deployment name
   • Confirm that deployment_name = "gpt-5.4-pro" exactly matches the deployment name configured in Azure OpenAI / AI Foundry, not just the base model name.
   • If the deployment name differs (for example, a custom name like gpt54prod), that exact name must be used in model=.
   3. Region/model support
   • Ensure that the specific model version deployed (for example, a gpt-5.x variant) is actually available and supported in the region where the Azure OpenAI resource is created.
   • If the model or version is not supported in that region, calls can fail even if the deployment appears present.
   4. Operation type vs. model capability
   • Verify that the deployed model supports the chat.completions API. If the deployment is of a type that only supports another API surface (for example, Responses API or a different task type), calling it via chat.completions.create can result in an "unsupported operation" error.
4. Error-handling and diagnostics
   • Since this is a 400 BadRequestError, retries will not resolve it; the request must be corrected.
   • Retrieve and log the request ID for support/troubleshooting:

          print(completion._request_id)
     

     This corresponds to the x-request-id header and can be used with Azure support.
5. Authentication and client configuration
   • The client instantiation pattern in the snippet matches the documented usage when using an API key:

          from openai import OpenAI
          
          client = OpenAI(
              api_key=api_key,
              base_url="https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1/",
          )
     

   • Alternatively, Microsoft Entra ID (DefaultAzureCredential + bearer token provider) can be used, but that does not change the semantics of a 400 error.

If, after verifying the endpoint, deployment name, model support in the region, and that the model is intended for chat.completions, the error persists, capturing the _request_id and engaging Azure support with that ID is the next step, as the service is clearly reachable but rejecting the specific operation as unsupported.

---

References:

• Azure OpenAI supported programming languages (programming-language-python)