Difference between "azure/eu/gpt-5.1", "azure/gpt-5.1-2025-11-13" and generally the many models in the cost map

[pdeubel]

Hi, I have deployed a GPT-5.1 with model version "2025-11-13" in an Azure Foundry in Sweden Central with the deployment type data zone standard. I wanted to add that model to our LiteLLM config but I am struggling to select the correct base_model so I can properly track quota, costs etc. In the current cost model map I found:

• azure/gpt-5.1
• azure/gpt-5.1-2025-11-13
• azure/eu/gpt-5.1
• azure/us/gpt-5.1

What is the difference between these base models? What would be the correct one for my gpt-5.1 in an EU region with data zone standard deployment? I guess "azure/eu/gpt-5.1"? Why is there no versioned one for the eu deployments?

Feedback: This is highly confusing, I struggled with this some time ago because there are also "azure/" and "azure_ai/" models (what is the difference there?)

Any tips / documentation / answers would be highly appreciated, thanks!

[pandego]

Great question - this naming is confusing at first.

For your case (Azure OpenAI deployment in Sweden Central, data zone standard), use:

• litellm_params.model: azure/<your_deployment_name> (this is always your Azure deployment name)
• model_info.base_model: azure/eu/gpt-5.1 (this picks the EU pricing entry for cost tracking)

How the keys differ:

• azure/gpt-5.1: generic Azure price bucket
• azure/gpt-5.1-2025-11-13: version-specific bucket (same family, dated)
• azure/eu/gpt-5.1 / azure/us/gpt-5.1: region-specific buckets for cost mapping
• azure_ai/...: Azure AI Inference models/provider path (different provider integration than Azure OpenAI)

If your billing/pricing is region-specific, azure/eu/gpt-5.1 is the safer base_model choice.

Example:

model_list:
  - model_name: my-gpt51
    litellm_params:
      model: azure/my-gpt51-deployment
      api_base: ${AZURE_API_BASE}
      api_key: ${AZURE_API_KEY}
      api_version: "2025-01-01-preview"
    model_info:
      base_model: azure/eu/gpt-5.1

If you want, I can open a docs PR with a short "which Azure base_model should I pick?" table to make this explicit.

[pdeubel]

Thanks for you answer, yes this would be great if you could extend the docs here. The buckets (versioning, region specific) make sense. What does not make sense to me yet is the difference between "azure/" and "azure_ai/" -> it is not really clear when to use one or the other. Is this because Azure moved from the "Azure OpenAI Service" resource to "Azure AI Foundry"? Also why do I have to use e.g. "azure_ai/claude-sonnet-4-6" for the Anthropic models through the Foundry instead of e.g. "azure/claude-sonnet-4-6"?