What each gateway charges on top of provider prices

All four pass provider token prices through, so the difference is the fee layer. OpenRouter's pricing page lists a platform fee of 5.5% on the Standard plan and 8% on the Business plan, with fee discounts available on Enterprise. The Free plan lists no platform fee but is limited to 50 requests per day.

Cloudflare's Unified Billing docs say a 5% fee applies to all credits purchased, so a $100 purchase is a $105 charge, and inference is passed through at the provider's per-token rate. Vercel's AI Gateway docs say it charges no markup and no platform fee on tokens, including with your own key, and add that you are responsible for any payment processing fees. LiteLLM's proxy has no gateway fee because you run it yourself, which moves the cost to hosting and a Postgres database.

Fee math on one workload

Take $10,000 of provider spend in a month, paid through the gateway's prepaid credits, and read each fee as a percentage of that spend.

  • OpenRouter Standard at 5.5%: $550. OpenRouter Business at 8%: $800.
  • Cloudflare Unified Billing at 5%: $500.
  • Vercel AI Gateway at no markup: $0 in gateway fees, plus any payment processing fees.
  • LiteLLM self-hosted: $0 in gateway fees, plus the server and database you run.

Bring your own key changes the arithmetic

OpenRouter's pricing page lists BYOK on Standard and Business as $25,000 of list-price inference per month with no fees and a 5% fee after. At $10,000 of monthly spend on your own provider keys, the OpenRouter fee is $0. At $40,000, the first $25,000 is free and the 5% reading on the remaining $15,000 gives $750. Vercel lists BYOK on its paid tier with no markup or fee, and Cloudflare lets a stored provider key take precedence over Unified Billing credits.

Vercel's pricing page adds a catch worth reading: when a request with your credentials fails, the gateway retries it with system credentials and charges that fallback usage to your credits balance. Cloudflare has the same shape of risk, since a request with no applicable credentials falls through to Unified Billing, and it ships a Require provider credentials setting that returns an HTTP 400 instead.

What each one can stop, not just report

OpenRouter's pricing page shows Budgets and Spend Controls as No on the Free plan and Yes on Standard, Business, and Enterprise. Vercel documents budgets that cap spend for a team, a project, an API key, or a team member. Cloudflare documents spend limit rules on individual gateways, scoped by model, provider, or custom metadata such as user or team.

LiteLLM is where the detail matters most. Its docs say virtual keys and budgets need a Postgres database, and that on a deployment without one the global max_budget fails open: a warning is logged once at startup, but nothing blocks at request time. A self-hosted proxy that looks configured can therefore still serve requests past the limit. Whichever gateway you pick, test the limit by setting a $1 budget and sending traffic until it returns an error.

Add-on charges also sit outside the headline fee. Vercel lists custom reporting at $0.075 per 1,000 tag, user ID, or quota entity ID writes and $5 per 1,000 reporting queries, and a team-wide provider allowlist at $0.10 per 1,000 successful requests on Pro and Enterprise. Per-request tagging for cost attribution is billed separately from the tokens.

Frequently asked questions

What is the cheapest OpenRouter alternative?

On 2026-10-14, Vercel AI Gateway lists no markup and no platform fee on tokens, and a self-hosted LiteLLM proxy has no gateway fee. Cheapest on the fee line is not cheapest overall: LiteLLM adds hosting and a database, and Vercel adds payment processing fees and per-use charges for some add-ons.

Does OpenRouter charge a markup on tokens?

OpenRouter's pricing page lists a platform fee of 5.5% on Standard and 8% on Business rather than a per-token markup line. Check the Billing and Pricing section of that page for how the fee applies to your purchase before you budget, because the fee base differs between credits and BYOK.

Can I enforce a hard spend limit on a self-hosted gateway?

With LiteLLM, only with a database connected. Its docs say a budget set without a database logs a startup warning and then does not block requests. Confirm enforcement with a low test budget before you rely on it in production.

Sources and further reading

  1. OpenRouter pricing (verified 2026-10-14)
  2. Vercel AI Gateway pricing (verified 2026-10-14)
  3. Cloudflare AI Gateway Unified Billing (verified 2026-10-14)
  4. LiteLLM proxy budgets and users (verified 2026-10-14)

FrugalAI uses primary documentation and published research where possible. Product capabilities and prices can change; verify vendor details before procurement or production changes.