Model version with AzureOpenAI

[miguel-arrf]

Hi!

I have an Azure OpenAI Service deployment where the model version is set to 2024-11-20.

If I want to switch to a different model version (e.g., 2024-08-06), how can I do that?

From what I understand so far, I need to create a new deployment — is that correct, or is there another way to switch versions?

Thank you!

[manimovassagh]

Yes, that's correct -- in Azure OpenAI, the model version is tied to the deployment. You can't change the model version on an existing deployment.

To switch versions, you have two options:

Option 1: Create a new deployment (recommended)

az cognitiveservices account deployment create \
  --name your-openai-resource \
  --resource-group your-rg \
  --deployment-name gpt4o-0806 \
  --model-format OpenAI \
  --model-name gpt-4o \
  --model-version "2024-08-06" \
  --sku-capacity 10 \
  --sku-name Standard

Then update your code to point to the new deployment name:

from openai import AzureOpenAI

client = AzureOpenAI(
    azure_endpoint="https://your-resource.openai.azure.com",
    api_key="...",
    api_version="2024-10-21",
)

response = client.chat.completions.create(
    model="gpt4o-0806",  # your new deployment name
    messages=[{"role": "user", "content": "Hello"}],
)

Option 2: Delete and recreate the deployment

If you don't want multiple deployments, delete the existing one and create a new one with the desired version. Be aware this causes downtime.

A few things to keep in mind:

• Not all model versions are available in all regions. Check Azure's model availability matrix.
• Older versions may get deprecated and auto-upgraded. Azure notifies you before this happens.
• You can run both deployments side by side to test before switching traffic.