Skip to main content
This article provides an overview of which regions support AI-assisted evaluators, the rate limits that apply to evaluation runs, how to configure virtual network support for network isolation, and using your own storage account to run evaluations.

​
Regional availability

​
Supported regions for Agent playground evaluations

The agent playground evaluations are supported in the following regions:

​
Supported regions for batch evaluations

The batch evaluations are supported in the following regions:

​
Supported regions for risk and safety evaluators

These regions support the following safety evaluators: Hate and unfairness, Sexual, Violent, Self-harm, Indirect attack, Code vulnerabilities, and Ungrounded attributes. Supported regions for Groundedness Pro:
  • East US 2
  • Sweden Central
Supported regions for Protected material:
  • East US 2

​
Supported regions for AI red teaming

AI red teaming is supported in the following regions.
  • East US 2
  • North Central US

​
Supported regions for data generation

The following regions support synthetic data generation and trace-to-dataset generation:

​
Azure OpenAI graders regional availability

For the Azure OpenAI graders regional list, see Regional availability.

​
Rate limits

The following rate limits apply to evaluation runs: Evaluation run creations are rate-limited at the tenant, subscription, and project levels. If you exceed the limit:
  • The response includes a retry-after header with the wait time.
  • The response body contains rate limit details.
Use exponential backoff when retrying failed requests.

​
Virtual network support for evaluation

For network isolation, you can bring your own virtual network for evaluation. To learn more, see How to configure a private link. Virtual network support for evaluation requires network injection (subnet delegation), but if you only need evaluation capabilities and do not require full agent support (Cosmos DB, AI Search, or project capability host), consider using the simplified evaluation-only setup template (15a) instead. It deploys a minimal network-secured environment tailored for evaluation scenarios with fewer resources and reduced complexity.
If you connect Application Insights, evaluation data is sent to it.
To prevent evaluation and red teaming run failures, assign the Foundry User role to the project’s Managed Identity during initial project setup.
The Foundry RBAC roles were recently renamed. Foundry User, Foundry Owner, Foundry Account Owner, and Foundry Project Manager were previously named Azure AI User, Azure AI Owner, Azure AI Account Owner, and Azure AI Project Manager. You might still see the previous names in some places while the rename rolls out. The role IDs and core permissions are unchanged by the rename.

​
Virtual network region support

Bringing your own virtual network for evaluation is supported in the following regions:

​
Bring your own storage

You can use your own storage account to run evaluations for your Foundry project, whether the project is configured with a virtual network or without one. For projects without a virtual network, you can use the storage account connection template. For projects with a virtual network, the storage setup is already included in the evaluation-only setup template (15a).
  1. For projects without a virtual network, create and connect your storage account to your Foundry project at the resource level. You can use a Bicep template, which provisions and connects a storage account to your Foundry project with key authentication.
  2. Make sure the connected storage account has access to all projects.
  3. If you connected your storage account by using Microsoft Entra ID, make sure to give managed identity Storage Blob Data Owner permissions to both your account and the Foundry project resource in the Azure portal.

​
Related content