[Get started](https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai-gateway%3FshowCreateKeyModal&utm_source=ai_gateway_landing_page&title=Get+an+API+Key) [Get started](https://vercel.com/signup?next=%2Fd%3Fto%3D%2F%5Bteam%5D%2F~%2Fai-gateway%3FshowCreateKeyModal%26utm_source%3Dai_gateway_landing_page%26title%3DGet%2Ban%2BAPI%2BKey) [Read the docs](https://vercel.com/docs/ai-gateway) # The AI Gateway for developers Hundreds of models, one API key, no markup. Text, image, video, audio. AI SDKChat CompletionsMessagesResponses ``` 1import { streamText } from 'ai' 2 3const result = streamText({ 4 model: 'xai/grok-4.6', 5 prompt: 'Why is the sky blue?' 6}) ``` [Read docs](https://vercel.com/docs/ai-gateway/sdks-and-apis/ai-sdk) SpaceXAI OpenAI Anthropic Gemini OSS [And More](/ai-gateway/models) 0% 0% No markup on tokensPay provider prices, never a cent more. Invoicing is available with no payment processing fees. See the [AI Gateway pricing docs](https://vercel.com/docs/ai-gateway/pricing). ## Optimize routing for availability, cost, or latency If a provider degrades, the Gateway fails over to the same model on another provider. Identical output, no downtime. Everyday requests route to a cost-efficient open model. Complex jobs escalate to a frontier model only when needed. Latency-sensitive requests route to the fastest-responding model for the lowest time to first token. Heavier requests fall through to a larger model. App Provider degraded? Claude Opus 5Anthropic Degraded Claude Opus 5Amazon Bedrock Latency figures are illustrative and vary by region, traffic, and prompt length. ## Routing, billing, and observability in one place Providing the developer experience and infrastructure to build, scale, and secure a faster, more personalized web. - [ One API key, hundreds of models. Unified billing and observability across your entire AI stack, with text, image, video, and audio models. ](/ai-gateway/models) - [ Route on behavior, fallback anytime. Automatic fallbacks during provider outages so your app stays up even when a model goes down. ](/docs/ai-gateway/models-and-providers/model-fallbacks) - OpenAI$12.47 Anthropic$8.22 SpaceXAI$4.31 Platform fee$0.00 Total$25.00 [ No markup, just fair prices. Pay exactly what providers charge with no platform fees. ](/docs/ai-gateway/pricing) ## Works with your existing AI stack `- X_AI_API_KEY``- ANTHROPIC_API_KEY``- OPEN_AI_API_KEY``+ AI_GATEWAY_API_KEY` Move existing OpenAI, Anthropic, and AI SDK integrations to AI Gateway with a base URL swap. [ AISDK The open-source AI toolkit designed to help developers build AI-powered applications and agents with React, Next.js, Vue, Svelte, Node.js, and more. ](/ai-sdk) [ OpenAI Chat Completions ](/docs/ai-gateway/sdks-and-apis/openai-chat-completions) [ OpenAI Responses ](/docs/ai-gateway/sdks-and-apis/responses) [ Anthropic Messages ](/docs/ai-gateway/sdks-and-apis/anthropic-messages-api) [ Open Responses ](/docs/ai-gateway/sdks-and-apis/openresponses) - [ Seamless migrationPoint your existing OpenAI or Anthropic SDK at AI Gateway. Same calls, no rewrites. ](/blog/cline-on-ai-gateway) - [ Bring your own keysBring your own provider keys with no platform fee. Existing commitments flow through. ](/docs/ai-gateway/authentication-and-byok/byok) - [ Models on day zeroImmediate launch partner with major labs. New models work the minute they ship. ](/ai-gateway/models) ## Recent ships [View the changelog](/changelog) - [ 13 AugustChangelog ### Gemini 3.7 Flash now available on AI Gateway for 50% off Joe McKenney, Jerilyn Zheng ](/changelog/gemini-3-7-flash-now-available-on-ai-gateway-for-50-off) - [ 12 AugustChangelog ### DeepSeek V4 Pro now runs updated weights on AI Gateway Jerilyn ZhengProduct, AI Gateway ](/changelog/deepseek-v4-pro-now-runs-updated-weights-on-ai-gateway) - [ 12 AugustChangelog ### Set up coding agents in one command with AI Gateway Sam Chitgopekar, Carlton Aikins, and 1 other ](/changelog/set-up-coding-agents-in-one-command-with-ai-gateway) > “Moving to the gateway is just so ergonomic. We get references to model names, and rely on Vercel to do the correct implementations and handle the edge cases.” Rob CheungCo-founder [How Zo Computer improved AI reliability 20×](/blog/how-zo-computer-improved-ai-reliability-20x-on-vercel) - 20x Improvement in AI reliability after switching - 30s To adopt a new model, down from an hour of code - 25% Improvement in average latency on model calls ## Security and compliance Route only to ZDR providers, no training or prompt logging, configurable per request or team-wide. - [ Zero Data RetentionEnforced on every request across your team. Routes only to providers under a ZDR agreement. ](/docs/ai-gateway/capabilities/zdr) - [ No training on your dataRoute only to providers that will not train on customer data, configurable per request. ](/docs/ai-gateway/capabilities/disallow-prompt-training) - [ Provider allowlistRestrict your team to approved providers. Enforced on every request, no code changes. ](/docs/ai-gateway/capabilities/provider-allowlist) ## Take control Manage usage across teams with budgets, quotas, and full request visibility. Set controls [ Budgets. Cap spend with budgets at the team, project, and API key level. ](/docs/ai-gateway/observability-and-spend/budgets) [ API key management. Create, view, and delete keys from the dashboard, CLI, or API. ](/docs/ai-gateway/authentication-and-byok/api-keys) [ Short-lived OIDC tokens. Authenticate with expiring credentials. No static credentials to manage. ](/docs/ai-gateway/authentication-and-byok#oidc-token-authentication) See everything [ Custom Reporting API. Tag requests by user, customer, feature, or environment. Analyze spend in the dashboard or your own systems. ](/docs/ai-gateway/observability-and-spend/custom-reporting) [ Dashboard observability. Usage, spend, requests, TTFT, and token counts at team, API key, and project scope. ](/docs/ai-gateway/observability-and-spend/observability) [ Request logs. Search and filter every request, follow traffic live, and open one to see how it routed and what it cost. ](/docs/ai-gateway/observability-and-spend/logs) Live on AWS MarketplaceBuy with private offers, using your existing cloud commits. [View the listing](https://aws.amazon.com/marketplace/search/results?searchTerms=vercel+ai+gateway) ## Every modality you need Text, image, video, realtime, speech, transcription, embeddings, and reranking through one endpoint. ### Text Access the latest from every major model lab and provider. Power your AI features all through a single endpoint. ### Image Generate and edit with the latest image models, no extra setup. ### Video New Ship production-ready video across a wide range of models from a single prompt. ### Realtime New Build voice and live multimodal experiences with realtime models through a single endpoint. ### Speech New Turn text into natural, expressive speech with the latest text-to-speech models, all through one endpoint. ### Transcription New Transcribe audio to text with leading speech-to-text models, no extra setup. ### Embeddings Vector embeddings for search, retrieval, and RAG pipelines, with every major provider available through one endpoint. ### Reranking Improve retrieval relevance by reordering results before they hit your model. OpenAI Vercel is a cloud platform for deploying and hosting websites, apps, and serverless functions with speed, scalability, and simplicity. It gives teams one workflow from development to global delivery, so products ship faster while keeping reliability and performance high across every request. With robust developer tooling and seamless integrations, Vercel enables engineering teams to collaborate efficiently and manage code from preview to production. Its edge network and automatic scaling help you serve users globally with minimal latency and maximum uptime, so you can focus on building great products while Vercel handles infrastructure, deployments, and optimizations to ensure a fast, consistent user experience at scale. \[0.234, -0.198, 0.567, 0.012, -0.823, 0.445, -0.671, 0.298, 0.789, -0.234, 0.617, 0.082, -0.301, 0.519, -0.448, 0.176, 0.902, -0.057, 0.388, -0.741, 0.263, 0.014, -0.598, 0.831, -0.122, 0.476, 0.249, -0.687, 0.354, 0.911, -0.205, 0.066, 0.498, -0.379, 0.732, -0.461, 0.187, 0.853, -0.092, 0.541, 0.318, -0.226, 0.679, -0.518, 0.034, 0.793, -0.347, 0.612, 0.158, -0.806, 0.273, 0.469, -0.135, 0.587, 0.821, -0.044, 0.396, -0.752, 0.218, 0.503, -0.661, 0.129, 0.874, -0.317, 0.452, 0.085, -0.539, 0.706, -0.198, 0.361, 0.927, -0.475, 0.244, -0.683, 0.518, 0.039, -0.792, 0.155, 0.642, -0.288, 0.471, 0.836, -0.107, 0.523, -0.366, 0.018, 0.748, -0.491, 0.265, 0.882, -0.146, 0.397, 0.609, -0.728, 0.184, 0.456, -0.029, 0.713, -0.554, 0.298, 0.067, -0.421, 0.836, -0.193, 0.512, 0.347, -0.768, 0.124, 0.658, -0.385, 0.901, 0.071, -0.469, 0.234, 0.587, -0.812, 0.156, 0.473, -0.628, 0.319, 0.052, -0.741, 0.486, 0.207, -0.563, 0.894, -0.135, 0.428, 0.671, -0.298, 0.519, 0.084, -0.756, 0.347, 0.918, -0.412, 0.176, -0.689, 0.245, 0.561, -0.328, 0.073, 0.842, -0.197, 0.453, -0.726, 0.288, 0.617, 0.039, -0.504, 0.871, -0.265, 0.392, 0.158, -0.673, 0.527, 0.084, -0.439, 0.916, -0.221, 0.358, 0.495, -0.782, 0.146, 0.629, -0.317, 0.058, 0.847, -0.473, 0.196, 0.534, -0.628, 0.279, 0.913, -0.045, 0.461, 0.184, -0.752, 0.398, 0.625, -0.171, 0.043, 0.789, -0.526, 0.314, 0.867, -0.082, 0.471, -0.638, 0.219, 0.582, -0.395, 0.146, 0.704, -0.273, 0.519, 0.038, -0.846, 0.187, 0.493, -0.561, 0.328, 0.075, -0.412, 0.901, -0.246\] - 1Cache invalidation guide - 2How streaming works - 3Routing edge cases - 4Provider failover patterns ### Text OpenAI ### Image Google ### Video New SpaceXAI ### Realtime New OpenAI ### Speech New ElevenLabs 0:00 0:00 ### Transcription New Deepgram You can build and host many different types of applications from static sites with your favorite framework, multi-tenant applications or micro-frontends to AI-powered agents. Deploy globally in seconds, scale automatically with traffic, and ship every change with preview deployments, observability, and built-in security on every request. ### Embeddings OpenAI ### Reranking Cohere ## Works with the tools your team already uses Route 11+ AI coding agents through AI Gateway with a base URL change. Get unified observability and spend tracking across every tool, no matter who built it. [See all supported coding agents](/docs/ai-gateway/coding-agents) Or skip the config files. The Vercel CLI detects the supported agents on your machine, provisions an API key, and writes their configuration for you. Terminal ``` vercel ai-gateway coding-agents setup ``` - [ Claude CodeAnthropic’s coding agent. Route it through AI Gateway’s Anthropic-compatible endpoint. Works with Claude Code Max too. ](https://vercel.com/docs/ai-gateway/coding-agents/claude-code) - [ OpenAI CodexOpenAI’s coding agent. Route it through AI Gateway’s Responses API so usage joins the rest of your AI spend. ](https://vercel.com/docs/ai-gateway/coding-agents/openai-codex) - [ OpenCodeOpen-source terminal coding agent with native AI Gateway support. Connect once, then switch between any model on the fly. ](https://vercel.com/docs/ai-gateway/coding-agents/opencode) - [ Blackbox AITerminal CLI for AI code generation and debugging, with access to every model in the catalog. ](https://vercel.com/docs/ai-gateway/coding-agents/blackbox) - [ ClineAutonomous coding agent for VS Code. Select Vercel AI Gateway as the provider for detailed token and cache metrics. ](https://vercel.com/docs/ai-gateway/coding-agents/cline) - [ Grok BuildSpaceXAI’s terminal coding agent. Point it at AI Gateway with two environment variables, and the model picker pulls the full catalog. ](https://vercel.com/docs/ai-gateway/coding-agents/grok-build) ## Get started This quickstart walks you through making your first text generation request with AI Gateway. [Read the quickstart](https://vercel.com/docs/ai-gateway/getting-started/text) Create a new directory and initialize a Node.js project. Terminal ``` mkdir ai-text-demo cd ai-text-demo pnpm init ``` Install the AI SDK and development dependencies. Terminal ``` npm install ai dotenv @types/node tsx typescript ``` Go to the [AI Gateway API Keys page](https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai-gateway%2Fapi-keys&title=AI+Gateway+API+Keys) in your Vercel dashboard and click **Create Key** to generate a new API Key. Create a `.env.local` file and save your API Key. Instead of using an API Key, you can use [OIDC tokens](/docs/ai-gateway/authentication-and-byok#oidc-token-authentication) to authenticate your requests. .env.local ``` AI_GATEWAY_API_KEY=your_ai_gateway_api_key ``` Create the `index.ts` file. index.ts ``` import { streamText } from 'ai'; import 'dotenv/config'; async function main() { const result = streamText({ model: 'openai/gpt-5.5', prompt: 'Invent a new holiday and describe its traditions.', }); for await (const textPart of result.textStream) { process.stdout.write(textPart); } console.log(); console.log('Token usage:', await result.usage); console.log('Finish reason:', await result.finishReason); } main().catch(console.error); ``` You should see the AI model’s response stream to your terminal. Terminal ``` pnpm tsx index.ts ``` - [ Provider and model routing with fallbacks ](https://vercel.com/docs/ai-gateway/models-and-providers/provider-options) - [ AI SDK documentation ](https://ai-sdk.dev/getting-started) - [ OpenAI Chat Completions API ](https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-chat-completions) Set up your project Terminal Install dependencies Terminal Set up your API key .env.local Create your script Create the `index.ts` file. index.ts Run your script Terminal ``` pnpm tsx index.ts ``` Next steps - [ AI SDK documentation OpenAI Chat Completions API ## Frequently asked questions How is AI Gateway priced? We offer tokens at list price from the upstream providers with no markup, including when you bring your own keys. Certain capabilities are available at higher plan tiers and metered separately. Invoicing is available with no payment processing fees. See the [pricing page](https://vercel.com/docs/ai-gateway/pricing) for details. What's the difference between using AI Gateway and going direct to each provider? AI Gateway gives you one integration, automatic failover, unified spend tracking, and one invoice across every major provider. Going direct means signing N contracts and stitching together N billing dashboards. Will AI Gateway work with our existing AI stack? Almost certainly. AI Gateway supports the [AI SDK](https://vercel.com/docs/ai-gateway/sdks-and-apis/ai-sdk), [OpenAI Chat Completions](https://vercel.com/docs/ai-gateway/sdks-and-apis/openai-chat-completions), [OpenAI Responses](https://vercel.com/docs/ai-gateway/sdks-and-apis/responses), [Anthropic Messages](https://vercel.com/docs/ai-gateway/sdks-and-apis/anthropic-messages-api), and an [OpenResponses](https://vercel.com/docs/ai-gateway/sdks-and-apis/openresponses)\-compatible endpoint. Migrating is typically a base URL swap with no code changes. Which modalities does AI Gateway support? [Text](https://vercel.com/docs/ai-gateway/getting-started/text), [image](https://vercel.com/docs/ai-gateway/capabilities/image-generation), [video](https://vercel.com/docs/ai-gateway/capabilities/video-generation), [embeddings](https://vercel.com/docs/ai-gateway/capabilities/embeddings), and [reranking](https://vercel.com/docs/ai-gateway/capabilities/reranking), all through the same endpoint. Browse the [full model catalog](https://vercel.com/ai-gateway/models) for specific models and providers. How does AI Gateway handle our enterprise security and compliance requirements? AI Gateway supports Zero Data Retention routing, a no-training guarantee, and team-wide provider allowlists. See the [security overview](https://vercel.com/docs/ai-gateway/capabilities) for full details. What observability does AI Gateway provide out of the box? A [dashboard](https://vercel.com/docs/ai-gateway/capabilities/observability) with usage, spend, request volume, TTFT, and token counts, broken down by model, provider, and project. For deeper analysis, the [Custom Reporting API](https://vercel.com/docs/ai-gateway/capabilities/custom-reporting) lets you pull the same data into your own tools. Can we use our existing provider contracts and committed spend? Yes, through [BYOK](https://vercel.com/docs/ai-gateway/authentication-and-byok/byok). Bring your own keys for almost every supported provider and your existing commitments flow through. We try BYOK first and only fall back to system credentials on failure. Do I pay per request or get invoiced? AI Gateway uses pre-purchased credits by default. Top up in the [dashboard](https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai-gateway&title=AI+Gateway+Dashboard) and usage is drawn down per request. Enterprise customers can switch to a single consolidated invoice from Vercel covering every provider in their routing pool. For invoicing, [reach out to sales](https://vercel.com/contact/sales) for more details. Can we purchase AI Gateway through AWS Marketplace? Yes. AI Gateway is live on [AWS Marketplace](https://aws.amazon.com/marketplace/search/results?searchTerms=vercel+ai+gateway) and available to purchase via AWS private offers. Procure it directly through AWS Marketplace and apply the spend toward your existing AWS cloud commits, within your existing budgets and procurement processes.