Azure OpenAI Realtime client_secrets API returns OperationNotSupported for valid deployment

Hello Pranit Awasthi,

Welcome to Microsoft Q&A and Thank you for reaching out.

The POST /openai/v1/realtime/client_secrets call is reaching the service successfully (authentication is valid), but the service is rejecting the request because the deployment being targeted is not recognized as a Realtime-capable deployment for the Realtime operation. The error text - realtime operation does not work with the specified model This is most commonly encountered when

1. The deployment is not actually a GPT Realtime model deployment (even if the deployment name exists and looks correct). Realtime endpoints only work with the GPT Realtime model families listed as supported for Realtime.
2. A non-supported region/resource is being used for Realtime, because Realtime model availability is region-dependent. The Realtime WebRTC doc lists supported regions for global deployments as East US 2 and Sweden Central.
3. The request is passing a model name when the endpoint expects a deployment name .If the deployment name is different from gpt-realtime-mini, the call can be routed incorrectly and result in operation/model mismatch errors.

As asked if the deployment needs to be created specifically from a Realtime-capable model family (for example, gpt-realtime / gpt-realtime-mini) - yes. Only the Realtime API supports specific GPT Realtime models/versions.

As asked if there any additional configuration steps required to enable Realtime support on a deployment - No special “enable” toggle is required beyond meeting the documented prerequisites

No special “enable” toggle is required beyond meeting the documented prerequisites. The documented prerequisites are:

• A resource created in a supported region, and
• A deployment of a GPT Realtime model in that supported region, and
• Using the GA endpoint format with /openai/v1 in the URL

For WebRTC ephemeral tokens specifically, please use the GA client secrets endpoint (…/openai/v1/realtime/client_secrets) as the mechanism to get the ephemeral token

As asked if this error indicates that the deployment is based on a model that does not support the Realtime API, even if the deployment name is valid - Yes. This error is consistent with the Realtime operation being invoked against a deployment that is not a supported Realtime model deployment or is deployed in a non-supported region for Realtime.

Please note that for Azure OpenAI, model should refer to the deployment name chosen during deployment. If the deployment name is not exactly gpt-realtime-mini, set "model" to the deployment name instead.

Please consider the following troubleshooting steps

1. Confirm the deployed base model is actually a GPT Realtime model
2. Confirm the resource region is supported for Realtime
3. Use the deployment name in the request payload
4. Keep GA endpoint format exactly as documented

References:

• Use the GPT Realtime API via WebRTC - Microsoft Foundry | Microsoft Learn
• Use the GPT Realtime API for speech and audio with Azure OpenAI - Microsoft Foundry | Microsoft Learn
• Migration from Preview to GA version of Realtime API - Microsoft Foundry | Microsoft Learn
• How to switch between OpenAI and Azure OpenAI endpoints with Python - Azure OpenAI Service | Microsoft Learn

Thank you!

Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the reply was helpful. This will be benefitting other community members who face the same issue.

Yes, the deployment must be created from a Realtime‑capable model family, and the error strongly indicates the deployment model does not support the Realtime API.

Key points based on the current Realtime GA behavior:

1. Realtime requires a Realtime model deployment
   • The client_secrets endpoint is only supported for deployments created from Realtime models:
     • gpt-4o-realtime-preview
     • gpt-4o-mini-realtime-preview
     • gpt-realtime
     • gpt-realtime-mini
     • gpt-realtime-mini (newer versions)
     • gpt-realtime-1.5
   • The session.model value in the client_secrets request must be the deployment name of one of these Realtime models, not a deployment of a standard chat/completions model.
   • If the deployment is based on a non‑Realtime model (for example, a regular gpt-4o or other chat model), the Realtime operations are not supported and an error like OperationNotSupported is expected.
2. No extra “Realtime enablement” beyond using a Realtime model
   • There are no additional flags or configuration steps to “turn on” Realtime for a deployment.
   • The only requirement is that the deployment itself is created from a Realtime‑capable base model listed above and is in a supported region.
3. What the error most likely means in this scenario
   • Given:
     • POST /openai/v1/realtime/client_secrets is used (GA endpoint format).
     • Authentication succeeds.
     • The deployment name exists and resolves.
   • The error message "The realtime operation does not work with the specified model." is consistent with:
     • The deployment being based on a non‑Realtime model family, even though the deployment name is valid.
4. Practical checks
   • Confirm in Azure AI Foundry / Azure OpenAI that the deployment is explicitly created from one of the Realtime models listed above.
   • Ensure the GA endpoint format is used (no api-version in the URL) and that the deployment is in a supported region for Realtime.

Once the deployment is recreated using a Realtime model (for example, a deployment of gpt-realtime-mini), the same client_secrets call with session.type = "realtime" and session.model = "<that deployment name>" should be accepted.

References:

• Use the GPT Realtime API for speech and audio (ai-foundry-portal)
• Use the GPT Realtime API for speech and audio (programming-language-python)
• Use the GPT Realtime API for speech and audio (programming-language-javascript)
• Use the GPT Realtime API for speech and audio (programming-language-csharp)
• Use the GPT Realtime API via WebRTC
• Use the GPT Realtime API via WebSockets
• Migration from Preview to GA version of Realtime API