Claude Opus 5 v2 keeps failing after a few minutes in Creating

Hi @Joachim Osenau ,

Thanks for the detailed troubleshooting. Based on the behavior you described, this does not look like a normal quota or deployment-template issue.

Microsoft documents several Claude-specific prerequisites in addition to model quota, including subscription eligibility, supported billing region, Azure Marketplace access, and permission to subscribe to the Anthropic offering.

Since you have already confirmed a paid PAYG subscription, available quota, accepted the Anthropic Marketplace terms, and reproduced the failure through both the Foundry portal and ARM, I would check the following before attempting another deployment:

Verify the Anthropic Marketplace agreement is Active for this subscription, rather than only confirming that the terms were accepted in the portal.

Verify the subscription is eligible for Claude based on its billing account/country and the selected deployment region.

Verify the required Marketplace/SaaS permissions and provider registration are present for the subscription. Microsoft documents the required Marketplace/SaaS permissions for partner model offerings.

Confirm that the selected Claude model/version and deployment type are currently available in the chosen region.

Your provisioningState = Failed with ResourceOperationFailure / InternalServerError and no additional error details does not expose the underlying provisioning failure through the portal. There are also recent Claude deployment reports with the same pattern where the failure occurs during the partner-model provisioning flow despite quota and regional capacity being available.

If the Marketplace/subscription eligibility checks above are all confirmed, please provide the failed deployment's correlation ID, operation ID, resource/deployment name, and UTC timestamp to Azure Support so that the backend provisioning operation can be traced.

I would not recommend continuing to delete and recreate the deployment, since you have already reproduced the same failure through both Portal and ARM.

Relevant documentation:

Deploy and use Claude models in Microsoft Foundry

Foundry Models from partners and community

Hi @Joachim Osenau

Thank you for reaching out to Microsoft Q&A. I am sorry you are experiencing this issue. I can relate, and I hope to point you in the right direction.

That said, based on the evidence, this does not look like a normal quota issue: GPT deploys successfully, while both Claude hosting options fail through the portal and ARM.

Claude deployments require these prerequisites. Claude prerequisites and partner-model permissions.

I would verify the following for the deploying identity:

• Required Azure Marketplace and Microsoft.SaaS permissions.
• Contributor or Owner access on the resource group.
• A supported deployment location for the selected Claude model/version.

If these checks are clean, open an Azure technical support request from the affected Foundry resource and include the correlation IDs, timestamps, resource ID, exact model/version and deployment type, plus the portal and ARM results. Ask support to inspect the failed deployment operation and the Marketplace/SaaS entitlement state.

The generic InternalServerError does not identify a specific backend cause, and I am not aware of a documented way to reveal it in the portal. Enable advanced diagnostics if permitted.

Create an Azure support request.

Please 'Upvote' (Thumbs-up) and 'Accept' as an answer if the response was helpful. This will help other community members facing the same issue.

Best regards, Andrew S Taylor

I think you just need to check whether the Anthropic offer in Azure Marketplace is active, and pass the correlation ID from the failed deployment to Azure Support. Foundry only gives you an InternalServerError in the portal — there’s nothing more you can pull from the UI. Claude can end up giving you an InternalServerError even when the Foundry quota looks fine, if the Anthropic offer in Azure Marketplace isn’t active.　

Hello @Joachim Osenau

Based on the tests you've already run, I don't think this looks like a normal quota or ARM template issue.

Microsoft's current Foundry documentation lists Claude Opus 5 v2 (Hosted on Azure) as supported for Global Standard deployments in West US 3, and also lists the PAYG quota at 40 RPM / 40K input TPM.

Claude models require a paid Azure subscription with an active pay-as-you-go billing method and an Azure Marketplace subscription to the Anthropic offering. Some subscription types aren't supported, including CSP subscriptions, free/student subscriptions, startup-credit-only subscriptions, and sponsored subscriptions using only Azure credits.

The generic ResourceOperationFailure / InternalServerError isn't enough to identify the underlying failure.

First, check one remaining item: Azure Marketplace subscription status for the Anthropic offer. Claude is sold through Azure Marketplace and uses Claude Consumption Units (CCU), so successful deployment requires more than simply having Foundry model quota available.

If the Anthropic Marketplace subscription shows as active and the billing account/subscription type meets the documented requirements, open an Azure Support request for Microsoft Foundry rather than continue recreating the deployment.

Provide Support with the failed deployment's UTC timestamp, deployment/resource ID, operation ID, and correlation ID from the Activity Log, and ask them to inspect the backend provisioning failure associated with the InternalServerError.

Don't conclude that this is an entitlement/risk flag yet. Unlike error 715-123420, which explicitly reports unusual account activity, the error you've posted contains no such indication. An InternalServerError can represent several backend provisioning failures, and only Microsoft can inspect the corresponding service-side operation.

If you can share the Activity Log JSON for one failed deployment, particularly the statusMessage, operationId, correlationId, timestamp, and resource type (with subscription/tenant identifiers removed), we can check whether Azure exposed anything more useful before you escalate.

References:

Deploy and use Claude models in Microsoft Foundry

Claude models in Microsoft Foundry - availability and quota

Foundry partner-model region availability

Deploy Microsoft Foundry Models

Help make this community better for everyone: if this answer resolved your issue, please accept it or upvote it. If not, share more details in a comment so we can continue the discussion and find the right solution.