Deploying Deepseek R1 as Serverless API with .models.ai.azure.com endpoint format

Question

I have a problem with creating a deployment of Deepseek R1 of "Serverless" as opposed to "Azure AI Services" type.

I need the endpoint in the format of deploymentname.eastus2.models.ai.azure.com NOT services.ai.azure.com/models****. In one Azure AI Hub/Project I'm able to deploy Deepseek-R1 at .eastus2.models.ai.azure.com endpoint, but in a different Azure AI Hub/Project, it always deploys with services.ai.azure.com/models.

Both environments are using the same East US 2 regions.

How can I make the Deepseek-R1 deployment ensuring the resulting Serverless API endpoint is in the format of ending with .models.ai.azure.com?

Answer 1

Hi @Autoize,

We have noticed that you rated an answer as not helpful. We appreciate your feedback and are committed to improving your experience with the Q&A.

Thanks for sharing your findings! It looks like the "Deploy models to Azure AI model inference service" toggle being enabled by default was causing the deployment to use the services.ai.azure.com format. Disabling it ensured that the endpoint follows the correct models.ai.azure.com structure.

Additionally, another way to achieve the models.ai.azure.com format is by deploying the model through Azure ML Studio under the Model catalog section. This also ensures that the endpoint is provisioned correctly under the Serverless model inference framework.

Let us know if you have any further questions!

---

If this answer is helpful, do click Accept Answer and Yes. Thank you for your support.

Answer 2

Hello Autoize !

Thank you for posting on Microsoft Learn.

The difference you're seeing in the endpoint format typically depends on how and where the model is deployed within Azure AI Studio, particularly:

• Azure AI Services (classic/custom deployment): uses services.ai.azure.com/models/...
• Azure AI Studio (Hub/Project with Serverless inference endpoint): uses *.models.ai.azure.com

If your deployment is done using Serverless Inference in a Project Workspace (Hub/Project), you need to create the Deployment via Project in Azure AI Studio (NOT directly in Azure AI Services pane)

• Go to Azure AI Studio (https://ai.azure.com)
• Navigate to your AI Hub → Open or create a Project
• Within the Project, go to Deployments → + Add Deployment
• Choose the Deepseek-R1 model and select “Serverless” deployment type
• Choose the East US 2 region
• After deployment, the endpoint will be in this format:

https://<deployment-name>.<region>.models.ai.azure.com

Deploying from Azure Portal or the Azure AI “Model Catalog” in the portal may route you to Azure AI Services directly, which creates endpoints in the services.ai.azure.com/models/... format.

Why the difference?

• The models.ai.azure.com endpoint format is for Serverless inference endpoints created within a project in Azure AI Studio.
• The services.ai.azure.com format is tied to traditional Azure AI Services or provisioned model endpoints (e.g., via Azure ML Studio or classic portal flows).

After deployment, go to: Azure AI Studio → Your AI Hub → Your Project → Deployments → Select Deployment

Under “Endpoint”, the expected format should be:

https://<your-deployment-name>.eastus2.models.ai.azure.com

Answer 3

No, this did not resolve the issue.
We are still encountering trouble creating an endpoint with the model.ai.azure.com format after navigating to https://ai.azure.com/, creating a new project, and then deploying the model.

First, we went to https://ai.azure.com/ and clicked "Create Project" and created a new project as you instructed.

Then, we browsed into that new project and navigated to the Models + Endpoints option on the sidebar. This menu option leads to a URL with the format https://ai.azure.com/build/deployments/model?...

Finally, we attempted deploying the model but it is still showing up with the services.ai.azure.com endpoint. To my understanding, this interface is the Azure AI Studio you are referring to, and not the Azure ML Studio or classic portal flows you referenced, correct?

Why is it still resulting in an endpoint of services.ai.azure.com and not models.ai.azure.com then?

Answer 4

It appears after switching the toggle to "off" for the preview feature "Deploy models to Azure AI model inference service" (which was set to "on" by default), we are now able to deploy an endpoint with the correct format.