Cloudflare AI Gateway alternatives: when to stay, and when to move
If your traffic already runs through Cloudflare, stay: AI Gateway's core features are free, it caches, logs, rate-limits and falls back between models, and its spend limits count your own-key traffic. Teams move when they need spend limits that hold under concurrent load, budgets and attribution across projects, teams and end customers, or routing for more than OpenAI-shaped chat; INFRO, Vercel AI Gateway, OpenRouter, LiteLLM and Portkey each fit one of those reasons.
A free core. Analytics, logging, caching and rate limiting at no charge; legacy logs are capped at 100,000 across all gateways on Workers Free and 10,000,000 per gateway on Workers Paid.
Your own keys, no fee. Stored keys (BYOK) carry no fee. Paying through Cloudflare's Unified Billing instead adds 5% on each purchase.
Spend limits that see BYOK. They count Unified Billing and BYOK requests for models with known pricing, and answer 429 at the limit.
Fallbacks and dynamic routing. The Universal endpoint falls back on error; dynamic routing builds flows with conditions, rate-limit and budget nodes, visually or as JSON.
For one team on Cloudflare, that is a complete answer.
Each option is described from its own pricing page and documentation. INFRO is listed first because this is INFRO's site; its limits are listed as plainly as everyone else's.
Teams that hold, or want to hold, their own provider contracts and need routing, failover, budgets and cost per customer across them.
How it charges
A flat plan of $35–$599 a month, chosen by request volume (100K–10M requests a month), never by spend. 0% markup on model spend: your providers bill you directly.
Strengths
Every request runs on a provider account you own; INFRO holds none, so nothing falls back onto someone else's contract.
Budgets refuse a request before the provider is called, per organization, project, member and key, and (from the Team plan) per end customer.
Each response names the provider, connection and model that served it, with an estimated provider cost.
Limits
No shared capacity: when every connection you own fails, the request fails with an error that says why.
Hosted only on its plans; a self-hosted deployment needs an enterprise agreement.
No prompt management or output evaluation.
Costs a subscription where Cloudflare's core is free.
Enterprises buying an AI gateway as part of a security platform, or teams self-hosting Portkey's open-source gateway.
How it charges
Production is $49 a month with 100,000 recorded logs, plus $9 per additional 100,000. The open-source gateway is MIT licensed. Enterprise is custom.
Strengths
Mature routing, fallbacks, guardrails and observability.
An MIT-licensed gateway you can host yourself.
Limits
Palo Alto Networks completed its acquisition on 29 May 2026; the product is now Prisma AIRS AI Gateway, so check its roadmap fits a team not buying the wider platform.
Logs, observability and prompt management are in the hosted app, not the open-source gateway.
They are not exclusive. A common shape keeps Cloudflare at the edge for what it does best (its network, caching, Workers AI) and puts budgets, attribution and audit across every provider account in a control plane behind it. Decide which layer owns retries and spend limits, so neither is applied twice.
Its core features are. Paying for model usage through its Unified Billing adds a 5% fee on each purchase; using your own provider keys carries no fee.
Do Cloudflare AI Gateway spend limits work with my own keys?
Yes, for models with known pricing. Cloudflare notes they are eventually consistent, so a burst of concurrent requests can briefly pass the limit.
Can I use Cloudflare AI Gateway and another gateway together?
Yes. Keep Cloudflare at the edge for caching and its network, and put budgets and attribution in a control plane behind it; decide which one retries.
Sources
Every statement about another company on this page comes from that company's own page, opened on the date shown. Prices and defaults change; check the source before you decide.
Vercel AI Gateway charges no markup and no platform fee on tokens; you pay the provider's list price.
Bringing your own provider key adds no Vercel fee.
When a request on your own credentials fails, Vercel retries it on its own system credentials and charges that usage to your Vercel AI Gateway balance.
Spend on your own credentials is not counted in Vercel AI Gateway budgets, so a budget cannot cap BYOK spend.
Past the allowance, BYOK usage is charged 5% of what the same model and provider would cost on OpenRouter.
By default, a BYOK request that fails or is rate limited on your key falls back to OpenRouter's shared endpoints, paid from your OpenRouter balance; a per-key "Shared capacity fallback" setting turns that off.
On 24 March 2026, litellm 1.82.7 and 1.82.8 on PyPI carried credential-stealing code for about 40 minutes; LiteLLM says the official Proxy Docker image was not affected and 1.83.0 is clean.
Palo Alto Networks completed its acquisition of Portkey on 29 May 2026.
The team building INFRO, a BYOK-first control plane for production AI. Articles are written from the engineering work on the gateway and cite their external sources.