One API for ~100 models

Vercel has opened alpha testing for the AI Gateway, a new product built on top of the AI SDK 5 alpha. The Gateway is designed to let developers switch between roughly 100 AI models without managing separate API keys, rate limits, or provider accounts for each one. Authentication, usage tracking, and future billing are all handled by the Gateway itself.

Requests are written as standard app routes or scripts, as shown below:

import { streamText } from 'ai'
const result = streamText({
  model: 'xai/grok-3', // defaults to Vercel AI Gateway
  prompt: 'Tell me the history of the San Francisco Mission-style burrito'
})

A terminal command kicks off the setup and points you toward the model library:

pnpm install ai@beta

Why centralize model access

The pace of model releases shows no sign of slowing. But each new state-of-the-art model typically comes with its own proprietary API, which has meant developers get locked into a single provider inside their application code. The goal of the Gateway is to remove that friction so teams can keep up with AI progress without juggling a dozen accounts and keys.

Production AI workloads also hit practical walls: capacity limits, rate limiting, and provider downtime. The Gateway is meant to provide load balancing across providers and automatic failover when one provider degrades or goes offline. With inference costs dropping and providers competing on price, the Gateway also makes it easier to chase those cost savings without rewriting code.

Vercel says the Gateway is an extension of the infrastructure approach used internally to scale v0 to millions of users, where the team rapidly switches and load balances between a mix of providers.

Relationship to the AI SDK

The AI SDK is a free, open-source abstraction layer for AI model APIs across text, image, and audio modalities. The AI Gateway is a separate Vercel product built on top of that SDK, keeping a high-cohesion, low-coupling design. The Gateway takes full advantage of SDK features including tool calling, function arguments, streaming, retries, attachments, and structured outputs.

Pricing and availability

During the alpha, Gateway usage is free, with rate limits tied to your Vercel plan tier—similar to the current AI SDK Playground limits. Vercel plans to introduce pay-as-you-go pricing at general availability, with model pricing reflecting provider market rates, updated regularly. Bring-your-own-key support is also under exploration.

On the roadmap:

  • Load balancing and model failover
  • Pay-as-you-go billing
  • Bring-your-own-key support
  • Unified logging, usage tracking, and observability
  • OpenAI-compatible API

Trying it out

The alpha is an early release for feedback. Demo applications are available for Next.js and Svelte, with model support details and usage examples at ai-sdk.dev/model-library. Note that the Vercel AI Gateway is not yet ready for production use or for migrating existing projects.