I’m troubleshooting repeated HTTP 500 errors from the Azure OpenAI Responses API on a gpt-5-mini test deployment. The API generates a short narrative for a Word report in a local Django application. The test sends only synthetic registration totals. An earlier live test generated the Word draft successfully.
Deployment and request
- Azure OpenAI resource region — East US
Deployment — gpt-5-mini, Global Standard, model version 2025-08-07
API route — POST /openai/v1/responses on the resource’s .openai.azure.com endpoint
The request sets store to false and max_output_tokens to 1200.
What I observed
The standard Word report downloads successfully. The AI Word export fails and Django returns HTTP 502 to the browser.
I called the same reporting function directly from the Django shell with a small synthetic input of 10 submissions, 7 activations, and 3 pending registrations. The Azure API repeatedly returned HTTP 500. The diagnostic captured a request ID but showed no Azure error code or message. The captured failure occurred on 15 November 2026 at approximately 16:50 UTC (18:50 SAST).
A separate test in the Microsoft Foundry Chat playground on the same deployment returned HTTP 400 and displayed a trace ID.
Azure Service Health showed no active service issues when I checked after the failure.
What is the best way to narrow down this HTTP 500 when the response provides no error details? Is there a recommended minimal Responses API request or a resource-level diagnostic I can run to isolate the problem? I’d also appreciate guidance on whether the Foundry playground HTTP 400 warrants a separate investigation.
I can share the Azure request ID and playground trace ID privately if needed.I’m troubleshooting repeated HTTP 500 errors from the Azure OpenAI Responses API on a gpt-5-mini test deployment. The API generates a short narrative for a Word report in a local Django application. The test sends only synthetic registration totals. An earlier live test generated the Word draft successfully.
Deployment and request
Azure OpenAI resource region — East US
Deployment — gpt-5-mini, Global Standard, model version 2025-08-07
API route — POST /openai/v1/responses on the resource’s .openai.azure.com endpoint
The request sets store to false and max_output_tokens to 1200.
What I observed
The standard Word report downloads successfully. The AI Word export fails and Django returns HTTP 502 to the browser.
I called the same reporting function directly from the Django shell with a small synthetic input of 10 submissions, 7 activations, and 3 pending registrations. The Azure API repeatedly returned HTTP 500. The diagnostic captured a request ID but showed no Azure error code or message. The captured failure occurred on 15 November 2026 at approximately 16:50 UTC (18:50 SAST).
A separate test in the Microsoft Foundry Chat playground on the same deployment returned HTTP 400 and displayed a trace ID.
Azure Service Health showed no active service issues when I checked after the failure.
What is the best way to narrow down this HTTP 500 when the response provides no error details? Is there a recommended minimal Responses API request or a resource-level diagnostic I can run to isolate the problem? I’d also appreciate guidance on whether the Foundry playground HTTP 400 warrants a separate investigation.