Compare gateways

Experiential vs Vercel AI Gateway

Both bill list price with no token markup, and Experiential adds native Gemini, an OpenRouter-compatible surface, and no per-request fees for ZDR, regions or your own keys.

このページはまだ翻訳されていません。英語版を表示しています。

Summary

Vercel AI Gateway is a hosted gateway with no token markup, budgets, provider fallbacks and ZDR controls, and first-class AI SDK and OIDC support. Several controls are paid add-ons: team-wide ZDR, the provider allowlist, regional endpoints, and BYOK without fallback. Experiential charges no per-request fee for ZDR, regions or your own keys; ZDR routing and new BYOK connections need Pro, and the provider allowlist is Enterprise.

import { generateText } from "ai";

const { text } = await generateText({
  model: "anthropic/claude-sonnet-5", // a plain string goes through Vercel
  prompt: "Hello",
  providerOptions: { gateway: { order: ["bedrock", "anthropic"] } },
});

Side by side

ExperientialVercel AI Gateway
HostingHostedHosted
APIsChat, Responses, Messages, Gemini generateContent, embeddings, batches, images, speech, transcription, videoAI SDK, Chat, Responses, Messages, embeddings, plus image, video, speech, realtime
Native Gemini APIYes, every modelNot documented
Price on tokensList price, no markupList price, no markup
Bring your own keyNo markup; first paid key needs ProNo markup; needs purchased credits; $0.20 per 1K requests with fallback off
Zero data retentionPro: org switch or per request, no fee; verdict per responsePer request free; team-wide $0.10 per 1K requests
Provider allowlistEnterprise$0.10 per 1K requests (Pro and Enterprise)
RegionsProcessing-region allowlistRegional endpoints, 5% of inference cost
FailoverOrdered waterfall; slow lanes demotedProvider and model fallbacks, provider order
Spend limitsPer-key caps, budgets, tier rate limitsPer team, project, key, member; BYOK spend excluded
Content guardrailsStandard pack or your own classifierNot documented
Response cacheNone (prompt caching passed through)None (prompt caching)

Where Experiential is better

  • Data controls with no per-request fee. Org-wide zero-retention routing (Pro) and no-training routing (every plan) add no fee per request, and every response carries its ZDR verdict. See Data controls.
  • Your own keys with no per-request fee. Vercel charges one when fallback to its keys is off.
  • Native Gemini and an OpenRouter-compatible surface beside Chat, Responses and Messages. See Compare: OpenRouter.
  • Content guardrails per key, with the standard pack or your own classifier. See Content guardrails.
  • Slow providers demoted automatically. See Routing and failover.

When Vercel may fit better

Your app is deployed on Vercel and you want OIDC with no key to manage, you build on the AI SDK's gateway provider, or you need realtime or reranking routes.

Migration notes

WhatVercel AI GatewayExperiential
Base URLhttps://ai-gateway.vercel.sh/v1 (Anthropic SDK: bare host)https://api.experientiallabs.ai/v1 (Anthropic SDK: https://api.experientiallabs.ai)
AuthAI_GATEWAY_API_KEY or VERCEL_OIDC_TOKENAuthorization: Bearer YOUR_API_KEY; no OIDC, so add EXPLABS_API_KEY to the Vercel project env
AI SDK plain stringsmodel: "creator/model" routes through VercelUse a provider (createOpenAICompatible) or set globalThis.AI_SDK_DEFAULT_PROVIDER
Model idscreator/modelPlain slug from GET /v1/models
providerOptions.gateway (order, only, sort)Provider routingRemove it; routing is automatic. Keep zeroDataRetention intent as provider: {"zdr": true}
models fallback listModel fallbacksKeep the first model
BYOKDashboard or per requestConnect accounts in Settings, Connections; per-request credentials have no equivalent

Switch with an agent

Paste this into your coding agent. It inventories the project first and asks before anything it cannot map.

POST/v1/chat/completions
curl https://ai-gateway.vercel.sh/v1/chat/completions \
  -H "Authorization: Bearer $AI_GATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/gpt-6-luna",
       "messages": [{"role": "user", "content": "Hello"}],
       "providerOptions": {"gateway": {"order": ["azure", "openai"]}}}'

Sources, checked 2026-11-19

  1. Vercel AI Gateway: APIs and modalities, routing, budgets (BYOK excluded), prompt caching, zero markup.
  2. Pricing: no markup, BYOK requirements and fee, ZDR, provider allowlist and regional hosting surcharges, batch jobs.
  3. Security and compliance: ZDR, prompt training, provider and model allowlists.
  4. Experiential: /llms.txt, GET /api/models.