Vercel AI Gateway vs OpenRouter: pricing, routing and keys compared
Ayo Adeniran · Oct 13, 2026 · 5 min read
Vercel AI Gateway and OpenRouter make the same basic offer: one API key, many models, each billed at the provider's list price. Both work with the OpenAI SDK. Both fail over between providers. If you squint, they look interchangeable.
They are not. They differ in where they make their money, how they treat your own provider keys, and how they decide which provider serves a call. Those three differences decide which one is cheaper and calmer for you.
The short version
- OpenRouter is the broadest catalog on one balance: 600+ models from 80+ providers. It charges a fee when you buy credits, and gives your own keys a large free allowance.
- Vercel AI Gateway charges no markup on tokens, including on your own keys, and fits naturally if you build with the AI SDK. Some controls are paid add-ons.
Side by side
| OpenRouter | Vercel AI Gateway | |
|---|---|---|
| Token prices | Provider list price, every plan | Provider list price, no markup |
| Fee | 5.5% on credit purchases (Standard), 8% (Business) | None on tokens; payment processing fees may apply |
| Free tier | 25+ free models; 50 requests a day, 1,000 after a $10 purchase | A monthly free credit on a subset of models, at lower rate limits |
| Your own keys | First $25,000 of list-price inference a month free, then 5% | No fee; needs purchased credits |
| Default routing | Weighs effective price, throughput, latency, rate limits and reliability | Routes across healthy providers, with provider order and model fallbacks |
| APIs | OpenAI-compatible | AI SDK, OpenAI Chat Completions and Responses, Anthropic Messages |
| Spend controls | Budgets and spend controls on Standard | Budgets per team, project, API key or member |
| Data handling | zdr and data_collection per request |
No prompt or response retained; per-request ZDR on Pro and Enterprise |
| Tracing out | Activity logs and export | Trace drains to your OpenTelemetry tool, billed per trace |
Pricing: where the money is made
OpenRouter bills inference at list price and takes its fee when you top up: 5.5% on Standard, 8% on Business. Enterprise plans negotiate the fee down.
Vercel takes nothing on tokens, though you pay any payment processing fees, and some controls carry their own meters. Team-wide zero data retention costs $0.10 per 1,000 requests, a team-wide provider allowlist $0.10 per 1,000 successful requests, and trace drains $0.05 per 1,000 traces plus $0.50 per GB of trace data. Per-request ZDR and the per-request only filter cost nothing extra. Purchased Vercel credits expire one year after purchase.
For plain token spend, Vercel saves you OpenRouter's fee. Once you switch on team-wide controls and trace export, the gap narrows, by an amount that depends on your request volume.
Your own provider keys
This is where the two diverge most.
OpenRouter treats your keys as a first-class routing layer. Keys are tried before its shared endpoints, in an order you set, and you can save several per provider. Each key can be told never to fall back to shared capacity, for the models it covers or for its whole provider, and filtered by model, API key or workspace member. The first $25,000 of list-price inference a month on your keys is free; after that the fee is 5%.
Vercel charges no fee on your keys, but they require purchased credits, and when a request on your key fails, Vercel retries it on its own credentials and bills your balance. Budgets do not count spend on your keys, so Vercel's docs advise setting spending controls with the provider too.
If you need a call to run only on your contract, OpenRouter's "never use shared capacity" rules say so directly. If you want zero fees and are comfortable with fallback to Vercel's credentials, Vercel is simpler.
Routing
OpenRouter's default routing scores providers on effective price, throughput, time to first token, rate limits, reliability and tool-call quality, then spreads traffic across the best of them. Follow-up requests in a conversation stay with the same provider while its prompt cache is warm. You can sort by price, throughput or latency instead.
Vercel routes a model across healthy providers, lets you set a provider order and model fallbacks, and publishes measured uptime, latency and throughput per model and provider, so you can check the numbers yourself.
Which one fits
Pick OpenRouter for the widest choice of models on one balance, rich provider controls, and a heavy own-key workload that fits inside the $25,000 monthly allowance.
Pick Vercel AI Gateway if you build with the AI SDK or Anthropic's Messages API, want list price with no fee on tokens, and like budgets per project and per member.
A third option: Momor Gateway
Momor Gateway is worth a look if you hold your own provider contracts and care most about what happens when a provider fails.
- Same SDK, same routing fields. The OpenAI SDK works with a new base URL and key, and
provider.orderandprovider.ignorebehave the way OpenRouter users expect. Anthropic, Gemini and Vertex requests are translated; everything else passes through as you wrote it. - Your keys, with OpenRouter-style rules. Prioritized keys go before Momor's keys and fallback keys after; each prioritized key decides whether Momor's keys may stand in for it, and any key can be limited to models, API keys or members. A call on your key costs 5% of its list price from your credits.
- Failover that carries the conversation. A provider failing before it answers is stepped over. Reasoning, tool calls and results move with the conversation, and tool-call ids are rewritten into the form each provider takes. A stream that breaks mid-answer is never joined to another provider's answer: it stops by default, or, on Momor's own API, the next provider answers again from its first token.
- Zero data retention for a whole organization, with calls running only on keys marked as covered by an agreement, and call records that hold metadata only, deleted after 90 days.
- File processing for PDFs, Word documents, recordings and scans, before the model sees them.
What to weigh: the catalog is 12 providers and 108 current models; Momor's own keys serve only the providers Momor resells, so the rest run on your keys; credit purchases carry a 5.5% fee with a $0.80 minimum; and the 5% on your own keys starts with the first call.
Read the Gateway quickstart or the bring your own keys guide.