Azure Foundry Model Monitor sometimes presents 2x the requests?

Question

Hi,

During Azure AI cost analysis, we noticed sometime Azure Foundry Model Monitor is presenting 2x the request made to Azure AI (API), could this be linked to Azure Content Safety input/output analysis?

• Azure Foundry Model Monitor - 4,18k:

• Azure custom dashboard (Model Requests (Sum)) - 4,18k:

- Requests to Azure AI (external Azure dashboard) - 2,08k:

Thanks.

Answer 1

The documented metrics do not state that Azure Content Safety input/output analysis causes Model Requests to be counted twice.

What is supported is:

1. Model Requests counts calls made to the model inference API over a period of time.
2. In Azure Monitor, that metric can be filtered and split by dimensions such as ModelDeploymentName, ModelName, ModelVersion, and StatusCode.
3. If comparing dashboards, a mismatch is often caused by different scope, filters, dimensions, or time range.

To validate the discrepancy end to end:

1. In the Azure portal, go to Monitor > Metrics.
2. On Select scope, select the same resource used by the Foundry deployment view. If selecting at resource group or subscription level, ensure Resource types is set to Foundry Tools.
3. Select the Models metric category, and use Model Requests.
4. Add filters for the exact ModelDeploymentName and, if needed, split by StatusCode.
5. Confirm the time range matches the external dashboard exactly.
6. If the external dashboard is showing only successful API calls, compare that against Azure Monitor split by StatusCode, because Model Requests includes calls to the model inference API over the selected period.

If the Foundry/Monitor view still shows about 2x after scope, deployment filter, status code split, and time range are aligned, the provided documentation does not identify Content Safety as the reason.

---

• Monitor model deployments in Microsoft Foundry Models
• Monitor model deployments in Microsoft Foundry Models
• Monitor model deployments in Microsoft Foundry Models