Does Databricks Unity Gateway support non-Databricks models like Claude or GPT?

Summary

• Yes, Unity Gateway integrates with external frontier models from providers like OpenAI (GPT), Anthropic (Claude), Amazon Bedrock, and Meta.
• External models are registered as Model Provider Services, which are governance-enabled Unity Catalog objects available to all users with access control and observability.
• You can adopt any model and switch between providers without re-integrating, with full token usage tracking, cost attribution, and rate limiting.
• Unity Gateway enables organization-wide governance and spend tracking with minimal latency overhead.

Does Databricks Unity Gateway support non-Databricks models like Claude or GPT?
Yes. Unity Gateway supports any external frontier model provider, including OpenAI (GPT), Anthropic (Claude), Amazon Bedrock, and Meta. You register these as Model Provider Services, which are Unity Catalog objects that every team can access through Unity Gateway with governance and observability built in. This means your teams can adopt any model on day one and switch between providers without re-integrating.

What is Unity Gateway?

Unity Gateway is Databricks' unified API gateway for large language models. It provides a single control point for routing requests to both Databricks-hosted models and external model providers. Every request routes through Unity Gateway, where governance policies (permissions, rate limits) are enforced and usage is tracked. This eliminates the need to manage API keys and integrations across multiple external services.

How to choose between external model providers

When deciding which external models to integrate, consider these key factors:

Capability	Why it matters
Model performance for your use case	Different models excel at different tasks (reasoning, code generation, summarization). Test your workloads to find the best fit.
Cost per token	External providers price differently. Unity Gateway tracks spend by provider, model, and user, so you can measure and optimize.
Availability and geographic region	Some providers have regional endpoints. Latency from your Databricks workspace to the provider's region affects response time.
Rate limits and quotas	Each provider has limits. Unity Gateway lets you set guardrails (per-user, per-app rate limits) to control spend and prevent runaway costs.
Data residency and compliance	Verify each provider meets your data residency and compliance requirements before registering the provider service.

How Databricks integrates external models

Integration happens in three steps:

1. Register a Model Provider Service: An admin creates a Model Provider Service in Unity Catalog, specifying the external provider (OpenAI, Anthropic, etc.) and supplying the API key or credentials. This is a one-time setup.
2. Grant access: Use standard GRANT and REVOKE permissions to control which teams and users can access the model through Unity Gateway. You can grant at the team level or to specific applications.
3. Route requests: Applications and users query the model through Unity Gateway using standard LLM APIs (REST or OpenAI-compatible SDKs). Every request is logged, token usage is tracked, and rate limits are enforced.

Why use external models with governance

Managing external models directly requires each team to hold their own API keys, apply their own rate limits, and track costs manually. This scales poorly:

• Fragmented access: Teams maintain separate integrations; admins lose visibility.
• No cost attribution: Spending is siloed by team; the organization can't optimize.
• Manual rate limiting: Runaway spending happens silently until the bill arrives.

Unity Gateway centralizes this. Every request is governed through Unity Catalog permissions, every token is counted and attributed to a user or app, and rate limits protect your budget. You see spend by provider, model, and consumer, enabling data-driven decisions about which models to adopt.

FAQs

Can I switch between external models without changing my application code?

Yes. Because the Model Provider Service is a Unity Catalog object, you can swap providers by updating the service configuration. Your application continues to call the same endpoint; Unity Gateway routes to the new provider. This avoids costly re-integration cycles.

What external models are available on Unity Gateway?

Unity Gateway supports any external provider you configure, including OpenAI (GPT models), Anthropic (Claude), Amazon Bedrock, and Meta. Databricks also hosts foundation models like Claude Fable 5.1, Gemini 3.8 Flash, and GPT-6 Astra as alternatives to external providers.

How do I track spend on external models?

Unity Gateway logs token usage and cost to the system.ai_gateway.external_model_spend system table, aggregated hourly. You can query this table to see spend by provider, model, user, team, or custom tags, and visualize it in the Usage Dashboard.

What if an external provider has an outage?

If one provider fails, you can update the Model Provider Service to route to a fallback provider, or route part of traffic to a backup. Unity Gateway does not provide automatic failover, so you should test your failover strategy in advance.