Vercel AI Gateway vs Cloudflare AI Gateway: which should you use?
Choose Vercel AI Gateway for coding agents and high-volume paid traffic: it adds no gateway fee on tokens and doesn't rate-limit the paid tier. Choose Cloudflare AI Gateway for apps on Cloudflare that want a free gateway with response caching in front of their own provider keys.
Vercel and Cloudflare side by side
| Vercel | Cloudflare | |
|---|---|---|
| Gateway fee on tokens | None; card processing fees apply to credits | None on your keys; 5% on Unified Billing credits |
| Credits | Expire one year after purchase; monthly free credit until you buy | Auto top-up; can briefly go negative |
| Rate limit | None from the gateway on the paid tier | 200 requests a minute per gateway on Unified Billing |
| Budgets | Per team, project, key or member; a soft cap, HTTP 402 | Per gateway by model, provider or metadata; eventually consistent, HTTP 429 |
| Routing | Provider order by cost, speed or throughput; rules and virtual models you set | Auto Router (free beta) and Dynamic Routing flows you build |
| Caching | Provider prompt caching handled automatically | Gateway response caching, free |
| Logs | Included, 36 days of lookback | Workers Logs: 200,000 a day free, 20 million a month on Workers Paid |
| EU | EU or US inference per request at the provider's regional price | Not compatible with Cloudflare's Regional Services |
| Claude Code | Dedicated endpoint, one-command setup, any gateway model, Claude subscription passthrough | Claude on Anthropic, Bedrock or Vertex |
| Codex | Dedicated /codex endpoint | OpenAI models only |
From Vercel's AI Gateway pricing, budgets and Claude Code docs and Cloudflare's pricing, limits and Claude Code docs, October 13, 2026.
What a coding-agent month costs
On bought credits Vercel is about 5% cheaper, before its card fees. Neither gateway lowers the token bill on its own: Vercel leaves model choice to rules you write, and Cloudflare's Auto Router doesn't accept the Anthropic format Claude Code sends.
| Month of coding-agent use | Vercel | Cloudflare (Unified Billing) | Standard Compute |
|---|---|---|---|
| Typical days, Claude Sonnet 5.5 | $71 | $74 | Standard, $89 ($95 budget) |
| Typical days, Claude Opus 5.5 | $141 | $148 | Pro, $249 ($269 budget) |
| Busy days, Claude Sonnet 5.5 | $579 | $608 | Pro Plus, $499 ($549 budget) |
Typical and busy are the median and 90th-percentile coding-agent day in our Agent Burn Index (September 13 to October 12, 2026; Claude Code and Codex not in the sample), over 20 working days. Vercel before card fees; on your own Anthropic key both are list price. Standard Compute plans stop at their budget; routing routine steps to efficient models makes it last longer.
Which to choose
- Choose Vercel AI Gateway for coding agents on bought credits, apps deployed on Vercel, Claude Code with other models or on your Claude subscription, and traffic that would pass 200 requests a minute.
- Choose Cloudflare AI Gateway for apps on Workers, free response caching and DLP, spend limits that fall back to a cheaper model, or a free gateway in front of keys you already have.
- Choose Standard Compute for coding agents on a fixed monthly price: compute included, a model chosen per request to stretch the budget, a plan per key for teams and agencies, and requests that stop at the budget.
The catch
Both budgets let some spend through: Vercel's check runs at the start of each request, so the one that crosses the limit completes, and Cloudflare's limits are eventually consistent under bursts. Set each budget about 10% below your real limit.
Where Standard Compute fits
Standard Compute works like either gateway from the agent's side, one base URL for Claude Code, Codex and OpenCode, and makes the cost decisions both leave to you: plans from $19 a month include the compute, routine steps go to efficient models, and requests stop at the plan's budget.
FAQ
Is Vercel AI Gateway cheaper than Cloudflare AI Gateway?
On bought credits, yes: Vercel adds no fee on tokens (card processing fees aside) and Cloudflare adds 5% to Unified Billing credits. On your own provider keys neither charges a fee.
Which has a rate limit?
Cloudflare's Unified Billing allows 200 requests per 60 seconds per gateway. Vercel says it does not rate limit paid-tier requests; provider limits still apply to both.
Which is better for Claude Code?
Vercel: a one-command setup, a dedicated /claude-code endpoint, any gateway model in /model, and Claude subscription passthrough. Cloudflare's setup works for Claude on Anthropic, Bedrock or Vertex.
Does either keep traffic in the EU?
Vercel can ask for EU or US inference per request at the provider's regional price. Cloudflare lists AI Gateway as not compatible with its Regional Services.