Databricks-hosted inference
Use Unity Gateway to run inference on frontier proprietary and open-weight models hosted by Databricks or by third-party providers. Choose a model and the serving capacity that fits your workload.
Models available through Databricks
Unity Gateway serves models hosted by Databricks and models hosted by third-party providers. Regardless of where a model is hosted, you are responsible for complying with the provider's applicable terms of use. See Applicable model terms.
The following models are hosted by Databricks. Each model name links to its specifications. For models hosted by their providers, see Models hosted by third-party providers.
Model assets in system.ai are listed globally. Seeing a model there does not mean it is available in your workspace. Availability depends on your workspace region, cross-geo settings, and model availability. Check the Unity Gateway UI for the corresponding model service and confirm that you have permission to query it.
Listing a model does not send customer data to it for inference. See Model availability by region for regional details.
DeepSeek V4 Flash (0731) is deprecated and is planned for retirement on November 5, 2026. See Deprecated and retired models for the recommended replacement model and guidance for how to migrate during deprecation.
DeepSeek V4.1 Flash will be retired on November 19, 2026. See Deprecated and retired models for the recommended replacement model and guidance for how to migrate during deprecation.
For deprecated models and recommended replacements, see Deprecated and retired models.
Models hosted by third-party providers
The following models are hosted by their providers, not by Databricks, and are served through Unity Gateway. Databricks provides the access path, authentication, and routing. The provider operates the model-hosting environment, and the provider's terms govern your use of the model.
Provider | Model | Hosting |
|---|---|---|
TypeSafe AI, Inc. | Provider-hosted, served through Unity Gateway |
Use a model
Your task | How to use the models |
|---|---|
Build an application or agent | Call a model through the Unity Gateway API. System-provided model services in |
Apply AI to your data | Use |
Use a coding agent | Connect your coding agent to models through Unity Gateway. See Supported coding agents. |
Supported models and model identifiers vary by interface. See the linked guides for requirements and examples.
System-provided model services
Databricks provides ready-to-use, pay-per-token model services in system.ai, such as system.ai.claude-opus-5. New services are added as foundation models become available.
- A system user owns them, and you cannot delete them.
- By default, only metastore administrators can modify them. A metastore administrator can delegate management by granting the
MANAGEprivilege.
To restrict access, see Discover and govern access to model services.
Choose serving capacity
Option | Recommended for | How it works |
|---|---|---|
Pay-per-token | Workloads with variable or intermittent usage. | Pay for the tokens you use without provisioning dedicated serving capacity. |
Workloads that need dedicated capacity with flexibility to adjust it as requirements change. | Pay for allocated capacity with no term commitment. | |
Business-critical applications or agents with predictable usage and sustained capacity needs. | Reserve dedicated capacity for a fixed term and pay for the full reservation, regardless of usage. |
Unity Gateway model services can route to provisioned-throughput destinations. See each option's documentation for eligible models and setup requirements.
For batch inference with ai_query, use the AI Functions guidance to select supported models and serving options. Model and regional support vary by serving option.
Connect other model providers
You can also use Unity Gateway with your own model provider account. Connect the model provider and govern access using your organization's credentials.
See Connect an external model provider.