# Azure OpenAI

> Connect an Azure OpenAI resource by its name, key and deployments.

OpenAI models deployed in your own Azure resource, billed to your Azure subscription. Where a workplace allows Azure but not OpenAI directly, this is the tile to use.

## Set it up

1. In the Azure portal, open your Azure OpenAI resource. You need its **name** (the part before `.openai.azure.com`), one of its **keys**, and the names of the **deployments** you made, with the model each one runs.
2. In **Settings → Model Hub → Providers**, select the **Azure OpenAI** tile.
3. Paste the key into **API key** and the name into **Resource name**. Use **Base URL** instead for a gateway or a region-specific address.
4. Leave **API version** empty for Azure's `v1` API, or enter a dated version your resource needs.
5. In **Deployments**, list each as `deployment=model`, separated by commas: `prod-gpt=gpt-6-luna, embed=text-embedding-3-small`. A deployment named exactly like its model can be written once.
6. Select **Test**.

## Models and prices

Azure does not list deployments through the key you enter, so the Hub knows only the ones you type. Each deployment is priced and described by its base model, using the same table as the [OpenAI](openai.md) page. A deployment whose base model is not in that table is listed without a price, and a call on it is refused as unpriced until the table catches up; Azure's own prices by region are not modeled.

## Reasoning

As OpenAI: the **Reasoning** option is sent as `reasoning_effort`, with *Off* as `none`. Because a deployment's name says nothing about its model, the Hub tells the API outright that the deployment reasons whenever a level is set.

## Web research

OpenAI's web search tool, where your deployment and region allow it. Where they don't, the call fails with Azure's own message.

## Related

- [OpenAI](openai.md)
- [Add a provider](../add-a-provider.md)