How much does Cloudflare AI Gateway cost?
Cloudflare AI Gateway itself is free. You pay for inference, either to your own providers or as Unified Billing credits with a 5% fee ($100 of credit costs $105), and logs on gateways created since September 24, 2026 follow Workers Logs pricing. For one monthly price with the compute included, Standard Compute plans start at $19.
What's free and what isn't
| Item | Price | Detail |
|---|---|---|
| Gateway | Free | Analytics, caching and rate limiting on every Cloudflare plan |
| Your own provider keys | No Cloudflare fee | Requests bill to your provider account |
| Unified Billing | 5% on credit purchases | Provider price passed through with no markup; OpenAI, Anthropic, Google, xAI, Groq and Workers AI |
| Logs, Workers Free | 200,000 a day, 3-day retention | For gateways created on or after September 24, 2026 |
| Logs, Workers Paid | 20 million a month, then $0.60 per million | 7-day retention; Cloudflare Observability pricing from December 1, 2026 |
| Logpush | 10 million a month, then $0.05 per million | Workers Paid |
| DLP | Free | Guardrails bill as Workers AI inference |
| Auto Router | Free while in beta | Picks among Anthropic, OpenAI and xAI models; open models only if you allow them |
From Cloudflare's AI Gateway pricing, Unified Billing, Workers Logs and the Auto Router launch post, October 13, 2026. Accounts with gateways from before September 24 keep the older log allowances.
A coding-agent month on Unified Billing
Pinned to one Claude model, a typical month costs less through Cloudflare and a busy month less on a Standard Compute plan. Cloudflare's new Auto Router can move the bill too, by picking a model per request, but it takes Chat Completions and Responses requests only, so Claude Code traffic on the Anthropic Messages API isn't auto-routed.
| Month of coding-agent use | Unified Billing (list + 5%) | Standard Compute plan |
|---|---|---|
| Typical days, Claude Sonnet 5.5 | $74 | Standard, $89 with a $95 budget |
| Typical days, Claude Opus 5.5 | $148 | Pro, $249 with a $269 budget |
| Busy days, Claude Sonnet 5.5 | $608 | Pro Plus, $499 with a $549 budget (about 19 busy days) |
| Busy days, Claude Opus 5.5 | $1,217 | Growth, $999 with a $1,119 budget (about 19 busy days) |
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. On your own Anthropic key the same months cost 5% less, paid to Anthropic. Standard Compute plans stop at their budget; routing routine steps to efficient models makes it last longer.
Limits that matter for agents
- Unified Billing is rate-limited to 200 requests per 60 seconds per gateway and answers 429 above it. Requests on your own provider keys don't count.
- Spend limits (since June 5, 2026) block with 429 once spend in a window reaches the limit, scoped by model, provider or metadata, up to 20 rules per gateway. Cloudflare calls them "eventually consistent": a burst of concurrent requests can briefly pass the limit.
- A spend limit can send traffic to a cheaper model instead of blocking, through a Dynamic Route you build.
- Credits can go negative in rare cases; Cloudflare charges the difference at the start of each month.
- On Standard Compute the plan's budget is the cap: once a key's spend reaches it, the next request gets HTTP 402, and a request already running finishes.
The catch
The gateway is free; every token still costs list price. Most of a coding agent's bill is the context it resends on every step, and nothing on this page changes that unless a cheaper model handles the routine steps.
Where Standard Compute fits
Standard Compute works like a hosted gateway with the cost work done for you: plans from $19 a month include the compute, and the router picks a cheaper model for routine steps on all three agent APIs, Anthropic Messages included. Teams already on Cloudflare that want free caching, analytics and DLP on keys they already pay for should stay there.
FAQ
Is Cloudflare AI Gateway free?
The gateway's core features are free on every Cloudflare plan. Inference isn't: you pay your providers directly, or buy Unified Billing credits with a 5% fee. Logs on new gateways follow Workers Logs pricing.
What does Cloudflare Unified Billing cost?
5% on every credit purchase: $100 of credits costs $105. Cloudflare passes the provider's inference price through with no markup.
What is the Cloudflare AI Gateway rate limit?
200 requests per 60 seconds per gateway for Unified Billing traffic, answered with HTTP 429 above that. Requests that use your own provider keys don't count toward it.
Does Cloudflare AI Gateway pick the cheapest model?
Its Auto Router, free in public beta since September 30, 2026, picks a model per request when you send model cloudflare/auto in the Chat Completions or Responses format. The default pool is Anthropic, OpenAI and xAI models; open models join only if you allow-list them.
Does Cloudflare AI Gateway keep data in the EU?
Cloudflare's Data Localization Suite lists AI Gateway as not compatible with Regional Services, so there is no EU-only processing option for the gateway itself.