Share via

Policy evaluation failure (InvalidResourceProperties) when deploying GPT-4.1-mini and other models in specific regions

Roopan P M (Persistent Systems Inc) 0 Reputation points Microsoft External Staff
2026-05-08T15:02:31.2133333+00:00

We are encountering the following error when deploying Azure OpenAI models via Bicep/ARM in certain regions:

Policy evaluation returned compliance: for model gpt-4.1-mini/2025-04-14 with error: (Code: InvalidResourceProperties)

Affected Models: gpt-4.1-mini (v2025-04-14) and other newer models

Behavior: Deployment fails in some regions while succeeding in others

Deployment Method: Bicep/ARM templates (Microsoft.CognitiveServices/accounts/deployments)

Impact: Blocking automated IaC deployments for Solution Accelerator templates across multiple regions

Expected Behavior:

Model deployment should succeed in all supported regions without policy compliance errors when the model/version is officially available.

Actual Behavior:

Azure Policy evaluation returns InvalidResourceProperties before the deployment completes. This appears to be a mismatch between the policy definition's expected

resource properties and the actual model deployment schema for newer model versions.

Questions:

  1. Is this a known issue with Azure Policy definitions not yet recognizing few model names/versions?
Azure OpenAI in Foundry Models

1 answer

Sort by: Most helpful
  1. AI answer

    2026-05-08T15:02:47.99+00:00
    1. The error pattern is consistent with a policy-definition mismatch rather than the model name/version itself being invalid. The troubleshooting guidance for built‑in model‑deployment policies shows that:
      • Policy failures of this kind typically occur when the policy’s expected resource properties or parameter values (for example, model asset IDs, publisher names, or parameter keys) do not match the actual model asset being deployed.
      • For model‑deployment policies, asset IDs and publisher names must match exactly and are case‑sensitive, and parameter names must match the policy definition (allowedPublishers, allowedAssetIds).
      • If a policy is written or parameterized against an older set of models, newer models such as gpt-4.1-mini/2025-04-14 can be blocked until the policy parameters are updated.
      From the available guidance, this is the expected behavior when the policy definition or its parameters are out of sync with the current model catalog, rather than a problem with the Azure OpenAI service itself. The recommended actions are:
      • Verify the policy assignment is using the correct parameter names and values by inspecting the definition (for example, via az policy definition show --name "<definition-id>").
      • Ensure the allowed model asset IDs and publishers in the policy match the current model entries from the model catalog.
      • Confirm the policy effect is Deny only where intended; otherwise use Audit while updating the allowed list.
    2. For regional availability of gpt-4.1-mini/2025-04-14 on Standard deployments, the model‑summary tables show:
      • Standard (regional) deployments: gpt-4.1-mini/2025-04-14 is not listed as available in the Standard deployment table excerpt provided. That table lists Standard availability for several models (including gpt-4.1, o4-mini, gpt-4o, gpt-4, gpt-35-turbo, etc.) but does not mark gpt-4.1-mini as available in the shown regions.
      • Chat completions / regional model availability: In the chat‑completions regional table, gpt-4.1-mini/2025-04-14 is shown as available (✔) in multiple regions such as germanywestcentral, japaneast, koreacentral, northcentralus, norwayeast, polandcentral, southafricanorth, southcentralus, southindia, swedencentral, switzerlandnorth, and uksouth.
      • Data zone provisioned managed: In the data‑zone provisioned managed table, gpt-4.1-mini/2025-04-14 is also shown as available (✔) in regions including eastus, eastus2, francecentral, germanywestcentral, northcentralus, polandcentral, southcentralus, swedencentral, westeurope, westus, and westus3.
      Based on these tables, gpt-4.1-mini/2025-04-14 is available in many regions for chat completions and data‑zone provisioned managed deployments, but it is not listed as available for Standard SKU deployments in the Standard deployment table excerpt. This explains why deployments may succeed in some configurations/regions and fail in others where Standard SKU is not supported for that model. For IaC templates targeting multiple regions and SKUs, the deployment logic needs to align with the model‑summary tables:
      • Only deploy gpt-4.1-mini/2025-04-14 as Standard where the Standard table explicitly lists it (none of the shown Standard rows do), or
      • Use the supported deployment types (for example, data‑zone provisioned managed or chat completions) in the regions where gpt-4.1-mini is marked as available.

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.