Vercel AI Gateway gateway details

Vercel's official AI Gateway, emphasizing unified model calls, failover, caching, and observability; ideal for apps already deployed on Vercel.

Best for: For frontend/full-stack teams, SaaS projects, and apps already using the Vercel AI SDK.

Current status

Reachable

Latency reference

760 ms

/models check

Platform model catalog

Last verified

2026-07-01

Status notes

Suits the Vercel AI SDK and frontend/full-stack projects; an engineering gateway rather than a prepaid relay.

Pricing mode

Settled via your Vercel account and model usage; the docs stress the gateway itself adds no markup.

Price tags

Standard price, High stability, Pay-as-you-go

Free quota

Vercel account credit and promotions may change; refer to the official dashboard.

Usually you create an API key in the platform console, then replace the SDK's baseURL, model name, and auth details per the docs. Test streaming, function calling, error codes, and rate limits before launch.

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AI_GATEWAY_API_KEY,
  baseURL: "see the platform docs"
});

const response = await client.chat.completions.create({
  model: "gpt-4.1-mini",
  messages: [{ role: "user", content: "Test one call through an AI gateway." }]
});

• Supported models: OpenAI, Claude, Gemini, DeepSeek, Grok, Qwen, Embedding.
• Capabilities: OpenAI-compatible, Streaming, Function calling, Embeddings, Log query.
• Documentation completeness: High; editorial stability score 4.6 / 5.

• More an engineering gateway than a China-facing prepaid relay; mainland direct connection and billing entity need separate evaluation.
• Third-party relay pricing, models, and availability can change, so re-check them periodically.

Is Vercel AI Gateway compatible with the OpenAI API?

This entry is marked as supporting the OpenAI-compatible format, but the exact base_url, model names, and parameters still follow the official docs.

Is Vercel AI Gateway ready for production?

Run a small-traffic load test, rate-limit test, and error-code test first, and confirm the billing rules; keep a backup provider for important workloads.