How much does Vercel AI Gateway cost?
Vercel AI Gateway charges each model's provider list price with no markup and no platform fee on tokens; you pay card processing fees when you buy credits, and purchased credits expire after a year. A coding agent on Claude Sonnet 5.5 costs about $71 a month there on typical working days and $579 when every day is busy. Standard Compute is the cost-optimized alternative for coding agents: a flat plan from $19 with included compute, and a router that picks an efficient model per request.
What you pay on Vercel AI Gateway
| Item | Price | Detail |
|---|---|---|
| Tokens | Provider list price | "No markup and no platform fee on tokens." Claude Sonnet 5.5 is $2 input and $10 output per million |
| Buying credits | Card processing fees | Vercel doesn't publish a percentage. Enterprise invoicing has none |
| Free tier | Monthly free credit | A subset of models with lower rate limits. It stops once you buy credits, and needs a payment method |
| Your own provider keys | No fee | Paid tier only, and you still need purchased credits |
| Team-wide zero data retention | $0.10 per 1,000 requests | Pro and Enterprise. Per-request ZDR is free |
| Team-wide provider allowlist | $0.10 per 1,000 requests | Pro and Enterprise |
| EU or US inference | Provider's regional rate | Shown on each model page, for example EU +10% on Claude and GPT models |
| Credit expiry | One year after purchase | Auto top-up is available and off by default |
From vercel.com/docs/ai-gateway (pricing, FAQ, getting started, regional inference) and the public models list, checked October 13, 2026.
What a coding-agent month costs
Pinned to one Claude model, a typical month costs less on Vercel AI Gateway and a busy month less on a Standard Compute plan, which covers about 19 of its 20 days. Zero markup removes the fee, not the token bill. A hundred typical coding-agent requests cost $5.33 on Claude Sonnet 5.5 and $0.41 on GPT-6 Luna at list price, so the model behind each step decides the month. Vercel's own routing guide says "the classifier is the part you own"; Standard Compute's router makes that call per request.
| Month of coding-agent use | Vercel AI Gateway (list price) | Standard Compute plan |
|---|---|---|
| Typical days, Claude Sonnet 5.5 | $71 | Standard, $89 with a $95 budget |
| Typical days, Claude Opus 5.5 | $141 | Pro, $249 with a $269 budget |
| Busy days, Claude Sonnet 5.5 | $579 | Pro Plus, $499 with a $549 budget (about 19 busy days) |
| Busy days, Claude Opus 5.5 | $1,159 | Growth, $999 with a $1,119 budget (about 19 busy days) |
| Typical days, GPT-6 Luna | $5 | Starter, $19 with a $20 budget |
Month = 20 working days, before card fees. Typical is the median active day in our Agent Burn Index, busy the 90th percentile, from coding agents such as OpenCode, Kilo and Roo on our gateway, September 13 to October 12, 2026; Claude Code and Codex were not in the sample. Standard Compute budgets assume every request on that one model; routing routine steps to efficient models makes a budget last longer.
Budgets are a soft cap
- Budgets can be set per team, project, API key or member, reset daily, weekly, monthly or never. Over budget, requests get HTTP 402.
- Vercel's docs: "A budget is a soft cap, not a hard limit. The check runs at the start of each request, so the request that crosses the limit still completes."
- Spend on your own provider keys is not counted unless you add it with the --include-byok flag, and a new key budget can take a minute or two to apply.
- Standard Compute checks the same way: once a key's spend reaches the plan budget, the next request gets HTTP 402 and a request already running finishes. The difference is that the budget is compute included in a fixed price, not a limit on a balance you keep refilling.
Rate limits
The free tier is rate-limited per model and answers 429 above it. Vercel says the paid tier removes the gateway's own limits, leaving only the upstream provider's, and doesn't publish fixed numbers.
The catch
No markup is not the same as low cost. Every step of a coding agent resends its context at list price, so a busy month on Claude Opus 5.5 passes $1,100 on any zero-markup gateway unless routine steps move to cheaper models, and on Vercel choosing those models is your job.
Where Standard Compute fits
Standard Compute is the cost-optimized option for coding agents: a flat plan from $19 a month with included compute that stops at its budget, smart routing that sends routine steps to efficient models, a plan per key for teams and agencies, Claude Code on /v1/messages and Codex on /v1/responses, and Europe or US rules on any plan.
FAQ
Does Vercel AI Gateway charge a markup?
No. Vercel charges provider list price with no markup and no platform fee on tokens, including with your own keys. You pay any card processing fees when you buy credits.
Is Vercel AI Gateway free?
Every team gets a monthly free credit on a subset of models with lower rate limits, which starts at your first request and needs a payment method on file. It stops applying once you buy credits.
Do Vercel AI Gateway credits expire?
Yes. Purchased credits expire one year after purchase, as stated in the purchase flow.
Is Vercel AI Gateway cheaper than OpenRouter?
Usually by a few percent: Vercel adds no platform fee, while OpenRouter charges 5.5% on card credit purchases. Both charge list price for tokens, so the model you run matters far more.
Can a Vercel AI Gateway budget stop overspending completely?
Not exactly. Vercel calls budgets a soft cap: the request that crosses the limit still completes, and spend on your own provider keys isn't counted unless you include it.