GPT-5.6-Luna deployment returning 503 InternalServerError: "no healthy upstream" in Azure AI Foundry Playground and API

Hello Henok Birru

Thank you for reporting this issue.

Based on the information provided, the error is occurring both in the Azure AI Foundry Playground and through direct API calls, including with a minimal prompt. This suggests that the issue is unlikely to be related to your application logic or prompt content and may instead be associated with the deployed model endpoint or the underlying service infrastructure.

The error message:

{

"error": {

"message": "no healthy upstream",

"type": "server_error"

}

}

typically indicates that the service was unable to route the request to a healthy backend serving instance for the deployment.

To help narrow down whether this is a deployment-specific issue or a broader service-side problem, could you please provide:

• The deployment name and deployment type (Global Standard, Data Zone Standard, etc.).
• A failed request ID (x-ms-request-id) and the approximate UTC timestamp of the failure.
• Whether creating a new deployment of the same model or testing an alternative model in the same Azure AI Foundry resource exhibits the same behaviour.

In the meantime, we recommend:

Verifying whether other model deployments in the same resource and region are working successfully.

Testing the deployment from both the Playground and API (which you have already done) to confirm consistent reproduction.

Checking the Azure Service Health dashboard for any active incidents affecting Azure OpenAI or Azure AI Foundry services in the region.

Once the above information is available, we can further investigate whether this is related to regional capacity, a deployment-level issue, or a backend service incident.

Thank you, and we look forward to your update.

Hello @Henok Birru

Since the same 503 "no healthy upstream" occurs in both the Foundry Playground and the API, with no configuration changes and even for a minimal prompt, this strongly points to an upstream model-serving/service issue rather than your application code.

I checked the public Azure status information, and there isn't currently a published broad Sweden Central Azure outage. However, that doesn't rule out a model-specific or deployment-specific incident. Similar 500/503 no healthy upstream failures have previously been reported for Foundry model deployments, including in Sweden Central.

Check Azure Service Health → Service issues for your subscription, since subscription/region-specific incidents may appear there before, or without appearing on the public status page.

If the deployment continues returning 503s, open an Azure support case and include the deployment/model name, Sweden Central region, first failure timestamp (~12:00 EET), request/correlation IDs from the response headers, and confirmation that both Playground and API reproduce it. Don't recreate the deployment yet, as that could remove useful evidence and may not help if the unhealthy upstream is on the service side.

Azure Status

At this point, Microsoft would need to confirm whether GPT-5.6-Luna in Sweden Central is experiencing a model-serving/capacity incident; the public status information alone isn't sufficient to make that determination.

Help make this community better for everyone: if this answer resolved your issue, please accept it or upvote it. If not, share more details in a comment so we can continue the discussion and find the right solution.