Both Vercel AI Gateway and OpenRouter are one API key in front of many models, both are hosted rather than self-hosted, and it's easy to assume they're interchangeable. They're not, and neither is built to answer a third question worth asking alongside them: once you're past routing and into governing who can call what, is either of these actually enough.
TL;DR
- Vercel AI Gateway is one endpoint (
ai-gateway.vercel.sh/v1) with zero markup on tokens even with your own keys, automatic same-model failover across providers, and deep AI SDK integration - the tradeoff is it's built around the Vercel ecosystem. - OpenRouter is platform-agnostic with 315+ models, a genuinely useful free tier, and a fee structure (5.5% on card top-ups, 5% BYOK above 1M requests/month) instead of Vercel's flat $5/month credit plus zero-markup model.
- TrueFoundry is the pick if you need RBAC, budget enforcement, guardrails, and MCP governance on top of routing - the thing neither Vercel AI Gateway nor OpenRouter does at all - though if you only need routing and billing, either of the other two is the simpler, sufficient choice.
What each one actually is
Vercel AI Gateway
Vercel AI Gateway routes through a single HTTP endpoint, with models referenced as creator/model-name strings - anthropic/claude-opus-4.8, for example - and in the AI SDK, passing that string as the model automatically routes the call through the gateway with no separate client. It includes embeddings alongside chat completions, spend monitoring, automatic retries, load balancing, and Bring Your Own Key. The standout feature is automatic failover to the same model on a different provider when one degrades, keeping output identical while switching which infrastructure serves it. Every team gets $5 a month in included credits from first use, and there's no markup on tokens, including with your own provider keys - you pay what the provider charges, full stop.
OpenRouter
It is a hosted aggregator: one API key and one credit balance in front of 315+ models, with an OpenAI-compatible API so switching models is a one-line change. The free tier gives 25+ free models at 20 requests/minute and 50 free-model requests a day, rising to 1,000/day once you've put at least $10 of credit on the account - a genuinely useful way to try things before committing. Pay-as-you-go is straightforward credit purchasing, and Enterprise adds SSO, SLAs, and negotiated support. OpenRouter doesn't mark up the model price itself, but there's a 5.5% platform fee on credit-card top-ups (with an $0.80 minimum) and a 5% bring-your-own-key fee on usage above 1M requests a month.
TrueFoundry
Truefoundry is the third option worth putting next to these two, and the one this post ranks first overall. Same starting point - one OpenAI-compatible endpoint in front of 1,000+ models - but with the governance layer neither Vercel AI Gateway nor OpenRouter has: RBAC scoped to users, teams, and virtual accounts, budgets that support an audit-only rollout before you switch on hard blocking, and PII/prompt-injection/content-moderation guardrails as configuration rather than something you assemble yourself. It also ships a native MCP Gateway with proper inbound/outbound auth handling and per-tool approval gates for destructive actions, which is a different category of feature than either of the other two offers at all. Deployment is managed, hybrid, or fully self-hosted in your own VPC, so data residency concerns that would rule out both hosted options entirely aren't a blocker here.
Head to head
| Vercel AI Gateway | OpenRouter | TrueFoundry | |
|---|---|---|---|
| Tied to a platform | Vercel / AI SDK | None - fully platform-agnostic | None - managed, hybrid, or self-hosted |
| Provider markup | 0%, including BYOK | 0% on model price; 5.5% card-fee, 5% BYOK fee above 1M req/month | No markup; priced separately from provider cost |
| Free tier | $5/month included credits | 25+ free models, 50-1000 free requests/day | No free hosted tier; self-hosted has no license cost |
| Same-model failover | Yes, across providers | Not a documented feature | Yes, with weighted/latency-based routing and canary rollouts |
| Zero data retention | Yes (OpenAI/Anthropic/Google), paid add-on team-wide | Not a core feature | Configurable per deployment posture |
| RBAC / per-team budgets | No | No | Yes, first-class |
| MCP server governance | No | No | Yes, native MCP Gateway |
Where each one is the right call
Pick Vercel AI Gateway if your app already ships through Vercel and the AI SDK is already your interface - it's closer to a config change than new infrastructure, and the zero-markup-even-with-BYOK pricing is genuinely competitive. One independent benchmark found the native Anthropic SDK about 15-20% faster than the gateway on small prompts, with that gap nearly disappearing at large context, which is worth knowing if latency on short requests specifically matters to you.
Pick OpenRouter if you want a hosted option that isn't tied to any particular deploy target, or if the free tier matters for prototyping before you've committed to a paid plan anywhere.
Pick TrueFoundry if you've outgrown "just route the request" and need to answer who's allowed to call what, how much a team can spend before it's blocked, and whether your agents' tool calls are audited - all without assembling that governance layer yourself on top of Vercel AI Gateway or OpenRouter, neither of which is built to provide it.
Where this ranking doesn't hold: if your actual need is simple API routing for a small team or a single app, TrueFoundry's governance surface is more than you need, and the honest answer is Vercel AI Gateway (if you're on Vercel) or OpenRouter (if you're not) - both are simpler, and neither makes you configure RBAC and budgets you don't have a use for yet.
Which of these three are you actually running, and did the platform tie-in, the fee structure, or the governance gap end up being the deciding factor?
Top comments (0)