Source: Vercel AI Gateway docs (overview, model fallbacks, budgets)

Vercel's hosted multi-provider gateway: OpenAI- and Anthropic-compatible endpoints, zero token markup, provider order plus model fallbacks with per-attempt metadata, and stacked soft-cap budgets.

Contents
  • What: official documentation (overview updated 2026-09-14; model fallbacks 2026-10-02; budgets 2026-09-10)
  • Publisher: Vercel
  • Date: read 2026-11-14
  • Original: Overview, Model fallbacks, Budgets

Summary

A hosted gateway at ai-gateway.vercel.sh, callable from any cloud via the AI SDK or the OpenAI Chat Completions, OpenAI Responses or Anthropic Messages formats. It centralises credentials, logs, cost tracking, budgets, prompt caching and failover.

Key claims

  • Adds zero markup to provider token prices, including with BYOK. Auth by API key, or OIDC from Vercel deployments.
  • providerOptions.gateway.models lists fallback models; order/only set provider preference per model. Each model is tried across its providers before the next model.
  • Response metadata modelAttempts lists every model and provider attempt with status code, timing and error, so the serving model is auditable.
  • caching: 'auto' applies provider prompt caching; full-response caching is left to the application.
  • Budgets at team, project, API key and user scope stack; any exceeded scope rejects with HTTP 402 quota_for_entity_exceeded. Budgets are soft caps checked at request start, so the crossing request completes. BYOK spend is not counted but is still blocked by an exceeded budget. Refresh periods: daily, weekly, monthly, none.

Feeds

LLM gateways, Model routing and fallbacks

Suggest changes →