We are encountering an InternalServerError (HTTP 500): “Unable to get resource information” when using Azure AI Foundry / Azure OpenAI models.

When the incident occurred, applications consuming LLMs received a generic 500 Internal Server Error, which can originate from multiple layers (application, hosting infrastructure, APIM, or the model service itself). Initially, this made it difficult to isolate whether the issue was with the application or the underlying model.

Only after testing directly from the Foundry portal by sending a prompt did it become clear that the failure was due to an APIM / model-side issue. However, without actively sending a prompt, there is currently no reliable way to detect such model-level outages. As already highlighted, relying on prompt-based checks is both reactive and costly, especially if done periodically across multiple models.

Given that a similar issue occurred again on 29th January, our question is:

Is there any out-of-the-box health, status, or monitoring API provided by Azure / Foundry to proactively track the availability and health of deployed LLM models (including APIM and model runtime), without sending inference prompts?

If not, are there any recommended Microsoft-supported patterns or signals (metrics, logs, status endpoints, or service health integrations) to detect model-side or APIM-side outages in a cost-effective and proactive manner?