Cloudflare AI Gateway Review (2026)

Free because the gateway is not the product

Core features are free on every Cloudflare plan - dashboard analytics, caching and rate limiting - requiring a Cloudflare account and one line of code.

And critically, there is no percentage markup on your tokens.

That is the structural difference against OpenRouter, which takes 5.5% of everything you spend, forever, with no volume discounts. At $50,000 a month of provider spend, that is $2,750 monthly to OpenRouter and nothing to Cloudflare.

The reason Cloudflare can do this is worth understanding rather than treating as a gift: the gateway is not the product. It is a way to make Cloudflare more central to your infrastructure, and they monetise elsewhere. That is genuinely a good deal for you - as long as you are comfortable with the relationship it deepens.

The meter is logs

Plan	Price	AI Gateway logs/month
Free	$0	100,000
Workers Paid	From $5/mo	1,000,000

Logs, not requests or tokens. That is the number to model, and it means consumption depends partly on how much you choose to retain rather than purely on traffic. Persistent logs are available on all plans with storage limits varying by plan.

For most teams 100,000 is a real free tier rather than a trial, and $5 for ten times that is close to negligible - a striking contrast with Langtail’s $99 entry tier or Parea’s $150.

Unified billing solves a procurement problem

Introduced in 2026, unified billing lets you pay for third-party model usage from OpenAI, Anthropic and Google AI Studio through a single Cloudflare invoice.

This sounds administrative and is more valuable than it reads. The friction it removes is not technical:

• Separate vendor relationships, each needing approval
• Separate payment methods and invoices to reconcile
• Separate spend to track across three dashboards

In a company of any size, adding a vendor is a multi-week process. Consolidating model spend onto an existing supplier relationship sidesteps that entirely, and gives you one place to see total spend.

Nothing else in this category offers it.

The edge argument

Cloudflare AI Gateway runs on Cloudflare’s global edge network, so the control plane sits close to your users rather than in a single region.

For a component that sits in the request path, that matters. Every gateway adds latency by definition; one running at the edge adds less of it for geographically distributed users than a single-region proxy does.

This is inherited capability rather than something built for AI, which is also why the caching and rate limiting are mature - they come from Cloudflare’s core business rather than being a first attempt.

Also included: request retry, model fallback, DLP scanning and guardrails.

Where it stops

Evaluation. Observability centres on request metrics, logs, usage analytics and cost visibility.

You will see what was called, how often, how fast, and what it cost. You will not see whether the answers were any good.

That is a reasonable scope boundary, but it means pairing this with Langfuse, Logfire or a dedicated eval framework - the same architecture LiteLLM implies. This is the routing and caching layer, not the quality layer.

Provider coverage is also narrower - OpenAI, Anthropic, Google, Replicate and Workers AI, against LiteLLM’s 100+ integrations and OpenRouter’s 500+ models. If breadth of model access is the point, this is not the tool.

One honest caveat on the free tier

Cloudflare has stated it will continue expanding the product, with some new features free as core features and others potentially part of a premium plan, announced as they become available.

That is a fair and transparent position rather than a warning sign. But it does mean today’s free scope is not a permanent commitment. If a specific capability is why you are choosing this, check whether it is core or newly added - new is where the premium risk sits.

Should you use it?

Use Cloudflare AI Gateway if you already run Cloudflare, you want caching, rate limiting and cost visibility with no markup, and you have separate tooling for evaluation.

Don’t use it if you need broad provider coverage, structured evaluation in the same product, or you are not otherwise a Cloudflare customer - adopting Cloudflare to get a gateway is a bigger decision than adopting LiteLLM.

Bottom line: the most generous free managed gateway available, with genuine architectural advantages at the edge and a billing consolidation nobody else offers. Model your log volume, pair it with real evaluation tooling, and it is very hard to beat on cost.

---

Pricing, log limits, provider coverage and the 2026 unified billing addition verified against Cloudflare’s own documentation and third-party comparisons on 3 August 2026. This is a researched directory entry - we have not yet instrumented this gateway with our reference application.