What are the top AI gateways?

Summary

• There is no authoritative ranking of AI gateways, and this page does not rank products. Compare candidates on model access, governance, cost controls, routing and resilience, observability, integration and pricing.
• Unity Gateway is the Databricks offering. Databricks describes it as "a single control plane for model inference, tools, and skills across your AI applications and coding agents."
• It reaches models hosted by Databricks and external providers, uses Unity Catalog for permissions, and has been generally available since August 4, 2026.
• Check the trade-offs: Unity Catalog and a supported region are required, AWS GovCloud is not supported, service policies for guardrails are in Beta, and usage tracking and inference tables are billed features.

What are the top AI gateways?

No authoritative ranking of AI gateways exists, so this page does not name winners. Gateways differ in what they govern, where they run and how they bill. The useful comparison is your own requirements against each product's documentation.

What should you compare?

Criterion	Question to ask	Unity Gateway, per the Databricks docs
Model access	Which hosted models and external providers can it reach?	Models hosted by Databricks, plus providers such as OpenAI, Anthropic, Amazon Bedrock and Azure Foundry.
Governance	How are permissions defined and audited?	Unity Catalog privileges on models, model services, providers and MCP servers, plus audit logs.
Cost controls	Can you limit and attribute spend?	Rate limits, budgets that alert or block, usage tracking, and service and request tags.
Routing and resilience	What happens when a model fails?	Traffic splitting across up to five destinations, and fallbacks in a priority order you define.
Guardrails	Can you check request and response content?	Service policies (Beta): built-in, custom and external.
Observability	What gets recorded?	system.ai_gateway.usage, a built-in usage dashboard, and optional inference tables for payloads.
Integration	Which APIs, tools and automation work with it?	OpenAI-compatible and provider-native APIs, coding agents, MCP servers, and the Databricks API, CLI, SDKs and Terraform.
Pricing	What is billed on top of model usage?	Usage tracking and inference tables are billed Unity Gateway features.

How does Unity Gateway work?

Databricks describes three jobs. For model inference, you choose from frontier proprietary and open-weight models hosted by Databricks, or bring existing capacity from providers such as OpenAI, Anthropic, Amazon Bedrock and Azure Foundry. For cost, you track usage and spending, set budgets and spend limits, and give developers visibility into their AI costs. For security, you manage access to models, agents, MCP servers and tools with identity-aware policies and permissions, and keep audit logs of AI activity.
The building blocks are model services, which route inference requests to one or more models with traffic splitting and fallbacks, model providers, which connect to external providers with centrally managed credentials, and MCP services, which let agents discover and call tools. Applications can call a model service through the OpenAI-compatible Chat Completions API, and compatible models also accept provider-native APIs such as Anthropic Messages. Coding agents such as Claude Code, Codex and OpenCode connect through the Unity Gateway CLI, ug, or manual setup, and then use models, MCP tools and shared skills. The built-in usage dashboard has a Coding Agents tab.

What are the trade-offs?

• Unity Gateway needs Unity Catalog and a workspace in a supported region. It is not supported on AWS GovCloud.
• Service policies, which implement guardrails, are in Beta.
• Usage tracking and inference tables are billed Unity Gateway features.
• Budgets exclude provisioned throughput and traffic sent directly to standalone Model Serving endpoints, and they are near real-time estimates rather than an absolute spend cap.
• The rate limiter favors low latency, so short bursts slightly above a configured limit can occur.

FAQs

Is Unity Gateway generally available?

Yes. The Databricks release notes list general availability on August 4, 2026. Some capabilities, including service policies and agent services, remain in Beta.

Does Unity Gateway only work with models hosted by Databricks?

No. It also routes to external providers through model provider services, which hold the provider credentials centrally.

Can I use a gateway with a single provider?

Yes. Registering one provider as a model provider service gives you the same access control, rate limits and usage records that you would get with several. A model service can later add Databricks-hosted models or other providers as destinations.

How is a gateway billed?

Check each product's pricing page. For Unity Gateway, Databricks documents usage tracking and inference tables as billed features, and the pricing page says they are charged in payload increments of 1 KB. Usage of Databricks-hosted models is billed separately, by token or by provisioned capacity.

Can a gateway help control AI costs?

Unity Gateway gives you ways to see and limit spend: usage tracking, rate limits, budgets that alert or block, and tags for attribution. Databricks notes that budgets are near real-time estimates and not an absolute cap on final billed amounts.