In November 2026, the best Cloudflare AI Gateway alternative depends on the constraint: Requesty or OpenRouter for broad model access, LiteLLM for self-hosted budgets, or Vercel AI Gateway for zero token markup. Cloudflare still earns a place on the shortlist because its core gateway features are free and it supports both fallbacks and per-user spend limits.
Which Cloudflare alternative fits your workload?
Cloudflare AI Gateway is a reasonable choice when its free core features and routing controls already cover your workload. The case for switching comes down to the constraint you hit: model access, hosting the gateway yourself, or budget enforcement.
| Gateway | When to consider it | Relevant capability |
|---|---|---|
| Cloudflare AI Gateway | You want free core gateway features | Analytics, caching, rate limiting, fallbacks and spend limits |
| Requesty | You want model access with managed routing and per-key caps | 600+ models, automatic failover and budget controls |
| OpenRouter | You want access to a broad model catalog | Hundreds of models; available on its own or as a provider through Cloudflare |
| LiteLLM | You want to run your own gateway with budgets | Self-hosted gateway with virtual keys and budgets |
| Vercel AI Gateway | You want zero token markup | Zero token markup |
For a wider comparison, see our guides to the best AI gateways and LiteLLM, Portkey, OpenRouter and Requesty.
Stay with Cloudflare when its free core gateway and controls cover your needs. Choose Requesty for broad model access with managed failover and per-key caps, OpenRouter to use its catalog on its own or through your existing Cloudflare gateway, LiteLLM for self-hosted budgets, or Vercel AI Gateway when zero token markup is the deciding factor.
What does Cloudflare AI Gateway cost?
Cloudflare includes dashboard analytics, caching and rate limiting for free on all plans. With Unified Billing, it adds a 5% fee to purchased credits and passes provider inference rates through without markup.
| Gateway or billing path | November 2026 gateway pricing |
|---|---|
| Cloudflare core features | Free on all Cloudflare plans |
| Cloudflare Unified Billing | 5% fee on credits purchased |
| Requesty | 5% on top of model costs, with all features included |
| OpenRouter Standard | 5.5% on card credit purchases, with a $0.80 minimum per purchase; 5% on crypto purchases; BYOK has no OpenRouter fee on the first $25,000 per month in list-price inference, then 5% on inference above that allowance |
| Vercel AI Gateway | Zero token markup |
Sources: Cloudflare pricing, Requesty pricing, OpenRouter pricing, the OpenRouter FAQ and Vercel AI Gateway pricing.
Take a workload with $10,000 in model costs, where the Cloudflare customer buys and uses exactly $10,000 in Unified Billing credits.
Cloudflare's credit fee would be $500, bringing the amount paid to $10,500. Requesty's 5% charge would also be $500, for a $10,500 total. The fee structures differ: Cloudflare charges when credits are purchased, while Requesty adds its fee on top of model costs.
Use the same $10,000 in model costs with your own provider keys on Cloudflare. Its core features and the Unified Billing credit fee add $0 in this example, compared with Requesty's $500 charge. That $0 is not the total Cloudflare gateway bill: log storage can add charges, and Guardrails is billed as Workers AI token inference.
For that $500 difference, compare the controls you would use, including Requesty's automatic failover chains and per-key caps. This example uses identical model costs; your bill will depend on the models and providers you choose.
Which log storage limit applies to your account?
Accounts that created their first gateway on or after September 24, 2026 use Workers Logs. Earlier accounts use Legacy Logs, according to Cloudflare's pricing page.
| Legacy Logs plan | Storage limit |
|---|---|
| Workers Free | 100,000 logs total |
| Workers Paid | 10 million logs per gateway |
At the limit, Legacy Logs stops saving new logs until you delete old ones, unless you turn on Automatic Log Deletion to remove the oldest logs. Check how close your account is to its limit before relying on the logs for troubleshooting.
For accounts using Workers Logs, these event allowances and retention periods apply through November 30, 2026. On Workers Paid, usage beyond the included events costs $0.60 per additional million events.
| Workers Logs plan | Included events | Retention |
|---|---|---|
| Workers Free | 200,000 per day | 3 days |
| Workers Paid | 20 million per month | 7 days |
Workers Logs moves to Cloudflare Observability pricing on December 1, 2026, with allowances based on data volume. The included allowance is shared across Workers logs, traces and AI Gateway logs. Free includes 0.5 GB of ingestion per day with seven-day retention; on the Free plan, new data stops being ingested once that daily 0.5 GB limit is reached. Paid includes 50 GB of ingestion and 10 GB-month of storage per billing cycle, with overages of $0.25 per GB ingested and $0.10 per GB-month stored.
Which models does Cloudflare AI Gateway support?
Cloudflare AI Gateway supports models through the providers listed in its provider documentation, including OpenRouter. If you already run Cloudflare, that gives you a way to try OpenRouter's model catalog through your existing gateway before deciding whether to switch.
Cloudflare has fallbacks, but routing still needs a plan
By default, Cloudflare triggers a fallback when a model request returns an error; a configured timeout can trigger one too. Dynamic Routing lets you build flows in a visual interface or JSON to evaluate conditions, enforce quotas and select models with fallbacks.
Retries are a separate control. Cloudflare supports up to five attempts for failed requests, with configurable delay and backoff, according to its request handling documentation. The Universal Endpoint is deprecated, so build new integrations on the REST API or Dynamic Routing.
If you want managed failover across providers, Requesty's routing includes automatic failover chains and weighted load balancing. Before switching, map the errors and timeouts that should trigger a fallback, then test the same cases against your current Cloudflare route.
Per-user budgets can overshoot during bursts
Cloudflare's spend limits can be scoped by model, provider or metadata such as a user or team. When Cloudflare Access protects the gateway, the authenticated user ID can be used to give each user an independent budget.
The limit is eventually consistent: a burst of concurrent requests can briefly exceed it before enforcement catches up. Cloudflare calculates each request's cost from its token usage and model pricing.
If your application assigns a separate budget to each customer, compare that behavior with Requesty's per-customer keys and budgets. Requesty has per-key caps and per-user spend limits; the useful test is how your own concurrent workload behaves near the boundary.
Watch the Unified Billing request limit
Cloudflare caps Unified Billing traffic at 200 requests per 60 seconds per gateway. Above that limit, requests using Cloudflare-managed credentials receive HTTP 429 with error code 2018. Requests using your own provider keys do not count toward the limit.
Before a launch or migration, check the busiest 60-second windows in your Unified Billing traffic. A daily request total can look comfortable while a short burst hits the gateway limit.
What to do before moving
Start with the constraint that sent you looking. For a spend limit, reproduce a burst near the budget boundary; for reliability, test an error and a timeout against your fallback route. For customer-level controls, map each customer to the key and budget you intend to use, as in our guide to per-customer keys and budgets.
Then compare the cost on your own model mix. Requesty's OpenAI-compatible gateway gives you access to 600+ models through one API, with provider failover and budget controls available on the same service.
Frequently asked questions
- What are the best Cloudflare AI Gateway alternatives?
- In November 2026, Requesty and OpenRouter are alternatives for broad model access, LiteLLM is an option for self-hosted budgets, and Vercel AI Gateway has zero token markup.
- How much does Cloudflare AI Gateway cost?
- Cloudflare AI Gateway's core features are free on all Cloudflare plans. Unified Billing adds a 5% fee when you purchase credits, so $100 of credit costs $105.
- Does Cloudflare AI Gateway support fallbacks?
- Yes. Cloudflare AI Gateway can trigger fallbacks on request errors or configured timeouts. Dynamic Routing can also choose models and apply fallback rules.
- Can Cloudflare AI Gateway set per-user budgets?
- Yes. Cloudflare AI Gateway can set independent per-user spend limits using the authenticated user ID from Cloudflare Access. Its spend limits are eventually consistent, so concurrent requests can briefly exceed a limit.
- What is the Cloudflare AI Gateway Unified Billing rate limit?
- Cloudflare AI Gateway limits Unified Billing traffic to 200 requests per 60 seconds per gateway. Requests using your own provider keys do not count toward that limit.
- How does Cloudflare AI Gateway compare with OpenRouter?
- OpenRouter can be used on its own or as a provider through Cloudflare AI Gateway. Cloudflare's core features are free and Unified Billing adds 5% to credit purchases; OpenRouter Standard charges 5.5% on card credit purchases with a $0.80 minimum, or 5% on crypto purchases, and its BYOK fee is 5% above $25,000 per month in list-price inference.
- OCT '26
Best AI Gateways in 2026: 10 Compared for Routing LLM Traffic
Compare the best AI gateways 2026 for hosting, pricing, routing and governance. Find a gateway that fits your workload, controls and monthly budget.
- JUN '26
Best LLM Gateway for Production (2026): LiteLLM vs Portkey vs OpenRouter vs Requesty
Choose the best LLM gateway for production: compare LiteLLM, Portkey, OpenRouter and Requesty on pricing, failover, security and deployment.
- OCT '26
Per-Customer LLM API Keys and Budgets for SaaS (2026)
Create a per customer LLM API key, set a tenant budget, track end users and turn model usage into SaaS invoices with a secure server-side setup.
