One API for Every Model

Assembling an AI application is fast work these days. With a stack like the AI SDK, a team can stand up a full frontend and backend in an afternoon: prompts go in, an LLM reasons, tool calls execute, and results stream back to the user. The hard part starts when that app has to survive real traffic.

Connecting straight to a single LLM provider is a fragile production strategy. Outages happen, rate limits bind, and every provider has its own key management and billing quirks. When the AI workload becomes mission-critical, the bottleneck is no longer integration—it's availability and consistent access to models.

Vercel's AI Gateway, now generally available, is built to handle that problem. It fronts requests to hundreds of models from dozens of providers through a single API, adds automatic failover, and removes the need to hold API keys or contracts with each vendor. The same infrastructure has been serving v0.app for millions of users.

Switching Models Should Be a One-Line Change

The AI landscape moves faster than integration cycles. Reasoning models became a standard capability almost overnight; tool use and the Model Context Protocol (MCP) went from experimental ideas to widely adopted standards in under a year. Teams that keep their system composable and avoid lock-in can move with that pace. Teams that hard-wire themselves to one provider's SDK and key management end up rebuilding plumbing every time a better model launches.

The AI SDK already normalized provider APIs to the point where swapping providers is a near-trivial edit. The AI Gateway extends that principle to infrastructure. With it, you can test a different provider by changing only the model string:

import { streamText } from 'ai'
const result = streamText({
  model: 'xai/grok-4', // defaults to Vercel AI Gateway
  prompt: 'How does Vercel AI Gateway have no markup on tokens?'
})

The Gateway, built on AI SDK 5, handles authentication, usage tracking, billing, and failover across supported models. No per-provider accounts, no juggling rate-limit policies, no stitching together multiple spend dashboards. That makes it a fit for teams that want to:

  • Evaluate or swap models dynamically as new ones ship
  • Aggregate rate limits across vendors rather than living within one provider's ceiling
  • Reach frontier models on launch day without a new integration
  • Avoid a single point of failure in their inference path
  • Track model usage and cost without hopping between vendor consoles

Built on the CDN's Reliability Playbook

Vercel's edge network already handles trillions of requests a year, and that infrastructure sits underneath the AI Gateway, keeping sub-20ms overhead. The architecture borrows from how CDNs solved availability for static assets: redundancy, failover, and optimization as a default, not an afterthought.

Pricing adds no markup on top of model list prices. You bring your own keys and contracts; the Gateway charges exactly the token rate the upstream provider sets, less payment processing fees. Vercel adds no rate limits of its own on top of the Gateway, though upstream provider limits still apply.

Getting Started

The AI Gateway is available today. New Vercel accounts receive $5 in credits every 30 days to try any model in the library—premium models included, with no usage restrictions. Note that the free credits stop once you make your first payment. See the full model list to check supported providers and chat with a model directly.