Azure OpenAI Documentation

Understanding Models vs Deployments

Key Difference

In Azure OpenAI, there’s an important distinction between available models and actual deployments:

  • client.models.list() returns the supported base models that your Azure OpenAI resource has access to (e.g., gpt-4, gpt-4o, gpt-5, etc.)
  • Deployments are the actual instances you’ve created to interact with these models

Important: You interact with models through deployments, not the base models directly.

Listing Your Deployments

There are three methods to view your Azure OpenAI deployments:

  • Navigate to your Azure OpenAI resource
  • Select Deployments under the “Management” section
  • View all active deployments with their configurations

2. Azure Management SDK (Python)

from azure.mgmt.cognitiveservices import CognitiveServicesManagementClient

# Use CognitiveServicesManagementClient
# Call client.deployments.list() with:
# - subscription_id
# - resource_group_name  
# - account_name

3. REST API

Use Azure’s REST API to programmatically retrieve deployment information for your Azure OpenAI resource.

Summary

  • client.models.list() → Shows what you could deploy
  • Portal/SDK/REST API → Shows what you have deployed

Deployment Limitations

One Model Per Deployment Rule

❌ Not Possible: You cannot create a single deployment (e.g., “GPT4TO5”) that uses both GPT-4 and GPT-5 simultaneously.

Reasons:

  1. Single Model Association: Each Azure OpenAI deployment is linked to one specific model version (e.g., gpt-4-0613, gpt-5-reasoning)

  2. Unique Deployment Names: Deployment names serve as unique identifiers for specific model instances within your Azure resource

Best Practice

Create separate deployments for each model you want to use, with descriptive names that indicate the model and purpose.