Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Foundation models are large, pre-trained models that you can query for chat, vision, audio, reasoning, and other generative AI tasks. This section describes how to discover available models, send query requests, and use provider-specific features on Azure Databricks.
Model services
Foundation models on Azure Databricks are served through model services—Unity Catalog securables that represent governed LLM endpoints. Azure Databricks provides a ready-to-use system-provided model service in the system.ai schema for each Databricks-hosted foundation model, such as system.ai.claude-opus-4-6. You invoke a model service by its fully qualified name.
Because a model service is a Unity Catalog securable, you govern access to it with standard Unity Catalog privileges, and you can query it from any workspace attached to the same metastore.
Your teams can also create custom model services that route across multiple models with fallbacks. See Create custom model services. To find the model services available to you, see Discover foundation models.
In this section
- Access generative AI and LLM models from Unity Catalog — Find the foundation models available to you.
- Use model services — Learn the options for writing and sending query requests.
- Query foundation models by type — Query chat, vision, audio and video, and reasoning models.
- Generate structured outputs — Constrain model output to a schema.
- Enable web search — Ground responses with real-time information from the web.
- Use provider native APIs — Use provider-specific API surfaces for OpenAI, Anthropic, and Google Gemini.
- Create custom model services — Create and share custom model services.
Supported models
For the foundation models you can query and their capabilities, see Discover foundation models.