Share via

Cannot deploy Anthropic Claude models - InvalidModelProviderData via REST API, UI deployment also fails

rongheng lin 40 Reputation points
2026-04-17T02:28:50.2033333+00:00
  1. Issue type: Technical
  2. Service: Cognitive Services
  3. Problem type: Deployment and setup issues
  4. Severity: C
  5. Support plan: Basic (free)

Attempts: 1. AI Foundry UI: deployment completes submission but resource shows Failed state with only "Internal Server Error" message. 2. REST API (api-version=2025-07-01-preview): persistent InvalidModelProviderData error regardless of where modelProviderData field is placed in request body. Marketplace offer is subscribed: anthropic/anthropic-claude-sonnet-4-6-offer/ anthropic-claude-sonnet-4-6-plan-new. Sold-to address on billing account is complete (organization name, country). Marketplace policy is set to Free + Paid on billing profile. Microsoft.SaaS resource provider is registered.

Azure Machine Learning
0 comments No comments

Answer accepted by question author

  1. Pilladi Padma Sai Manisha 7,720 Reputation points Microsoft External Staff Moderator
    2026-04-25T07:33:19.8866667+00:00

    Hi Rongheng,

    Thanks for the detailed description, that helps a lot. Based on what you’re seeing (portal showing Internal Server Error and REST returning InvalidModelProviderData), this isn’t a simple syntax issue in your request. It typically indicates that Azure AI Foundry is unable to validate or resolve the Anthropic Marketplace linkage behind the scenes.

    A few targeted checks and actions should help you move forward:

    The most important point is that Anthropic models (like Claude Sonnet 4.6) are not deployed by manually constructing modelProviderData. That field is internally populated by Azure when it successfully links your Azure AI resource to the Marketplace SaaS subscription. If you try to pass custom values (for example organization name or country), the service will reject the request with InvalidModelProviderData, which matches what you’re seeing.

    I recommend starting with the portal instead of REST to establish a clean linkage. In Azure AI Foundry, create the deployment directly from the model catalog. If the Marketplace subscription is correctly recognized, the platform will automatically bind the required provider data. If the same deployment fails in the UI with “Internal Server Error,” it usually confirms the issue is not your payload, but a backend validation or entitlement problem.

    Next, please verify that your Anthropic Marketplace subscription is properly connected to the same subscription and tenant as your Azure AI resource. You can check this in the Azure portal under SaaS resources. The subscription should be in an Active state. If there’s any doubt, re-subscribing to the offer often resolves hidden linkage issues by generating a fresh SaaS resource.

    Also ensure that you’re deploying in a region where Anthropic models are supported for Azure AI Foundry. Availability can vary, and deploying in an unsupported region can result in generic failures rather than clear validation errors.

    Since both UI and REST are failing in your case, I would strongly recommend checking the Activity Log for the failed deployment and capturing the correlation ID. If everything above looks correct, this is very likely a service-side issue, and providing that correlation ID in a support request will allow the product team to identify the exact backend failure.

    For reference, the following Microsoft documentation aligns with this setup and expected behavior:

    Azure AI Foundry model deployment guidance: https://learn.microsoft.com/azure/ai-foundry/how-to/deploy-models

    Azure Marketplace SaaS subscription and activation: https://learn.microsoft.com/azure/marketplace/purchase-saas-offer-in-azure-portal

    Azure AI Services REST API (preview reference): https://learn.microsoft.com/azure/ai-services/openai/reference-preview

    In short, avoid manually setting modelProviderData, confirm the Marketplace SaaS linkage, and use the portal to validate the deployment path. If it still fails, the correlation ID from Activity Logs will be key to resolving this quickly with Microsoft support.

    Hope this helps you get unblocked.

    Was this answer helpful?

    1 person found this answer helpful.

Answer accepted by question author

  1. Q&A Assist
    2026-04-17T02:29:07.6766667+00:00
    AI generated content. This question contains an answer created with AI from Q&A Assist. Learn more

    The InvalidModelProviderData error and failed UI deployment indicate a subscription/eligibility or region issue with Anthropic Claude in Azure AI Foundry, not a JSON placement problem.

    Key constraints to validate:

    1. Confirm subscription eligibility
      • Claude models in Foundry currently require Enterprise or MCA-E subscriptions. Other subscription types (including CSP, free trial, student, startup credit–based, sponsored/credits-only) are not supported.
      • If the current subscription is not Enterprise or MCA-E, Claude deployments will fail regardless of correct Marketplace subscription or REST payload.
      • From the documented errors: a "Subscription eligibility error" occurs when using a non-Enterprise or non-MCA-E subscription.
    2. Check unsupported subscription types The following are explicitly not supported for Claude in Foundry:
      • Enterprise accounts located in South Korea
      • Cloud Solution Provider (CSP) subscriptions
      • Subscriptions without an active pay-as-you-go billing method (for example, student, free trial, or startup credit–based accounts)
      • Sponsored subscriptions that only use Azure credits (if a credit card is on file, the card is charged instead of credits)
      If the subscription falls into any of these categories, deployments will fail even if:
      • The Anthropic Marketplace offer is subscribed.
      • Microsoft.SaaS is registered.
      • Billing profile and Marketplace policy look correct.
    3. Verify region and project setup
      • Claude models in Foundry must be used in a Microsoft Foundry project created in one of the supported regions: East US2 or Sweden Central.
      • If the Foundry project or resource group is in an unsupported region, deployments can fail with generic internal errors.
      • Ensure the deployment is a global standard deployment as per Claude model requirements.
    4. Validate Marketplace and permissions (already mostly done)
      • Marketplace subscription to the Anthropic offer is required and already in place.
      • Ensure the account has the permissions required to subscribe to model offerings and that Azure Marketplace access is allowed on the billing profile.
      • Confirm the user or service principal has Contributor or Owner on the resource group used for the deployment.
    5. Next steps to resolve
      1. Check the subscription type in Azure Portal (Billing → Subscriptions → select subscription → Properties) and confirm it is Enterprise or MCA-E and not CSP, free trial, student, or sponsored-only.
      2. If the subscription is not eligible, move the Foundry project and Claude deployment to an eligible Enterprise or MCA-E subscription.
      3. Ensure the Foundry project is in East US2 or Sweden Central and redeploy the Claude model there.
      4. If all eligibility and region constraints are satisfied and the error persists, delete the failed deployment and recreate it; similar issues have been resolved by recreating the deployment after some time.

    Once the subscription type and region match the documented requirements, the InvalidModelProviderData and internal error during deployment should be resolved without needing to change the REST body structure.


    References:

    Was this answer helpful?

    1 person found this answer helpful.

1 additional answer

Sort by: Most helpful
  1. rongheng lin 40 Reputation points
    2026-04-17T05:46:27.2366667+00:00

    The account is MCA-A (PAYG) . Accroding to the rules, it might be ok with claude models. But there is always "failed" after creating.

    Was this answer helpful?

    1 person found this answer helpful.
    0 comments No comments

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.