Integrate Azure OpenAI for text and vision generation

This promptfoo example demonstrates using Azure OpenAI with promptfoo for both text generation and vision capabilities. It covers text models like gpt-5.1 and gpt-4o, and vision-capable models such as gpt-4o, and includes separate configs for basic text evaluation and vision evaluation.

Use this example when you want to run promptfoo evaluations against your own Azure OpenAI deployments, whether for plain text generation or for vision tasks that need image input. It requires an Azure account with Azure OpenAI Service access, deployed models matching the ones referenced in the config, and your Azure OpenAI endpoint URL - the config files need their apiHost and deployment names updated to match your actual Azure setup before running.

Run basic text generation with the default config, or vision evaluation with a dedicated config file:

npx promptfoo@latest eval
### or
npx promptfoo@latest eval -c promptfooconfig.yaml

### Vision Models
npx promptfoo@latest eval -c promptfooconfig.vision.yaml

The vision example demonstrates three distinct ways to supply an image to a vision model: a direct URL to a web-hosted image, a local file referenced via a file:// path (automatically converted to base64), and a pre-encoded base64 data URI.

Set AZURE_API_KEY (or the alternative AZURE_OPENAI_API_KEY) in your environment or a .env file. If you get a 401 error, check that the key is set correctly, that the endpoint URL omits the https:// prefix, and that the deployment actually supports the capability you're requesting (e.g. vision). Full details are in promptfoo's Azure OpenAI provider docs and Microsoft's Azure OpenAI and vision documentation.

Teams evaluating Azure OpenAI deployments - text or vision - with promptfoo, who want working example configs for both capability types plus all three image-input methods already set up.