Gateway pricing
—
gateway overhead / month—gateway total
—effective markup
—overhead / year
Overhead as you scale
A flat fee shrinks as a share of the bill with volume; a percentage markup grows in step with it. Watch where the overhead crosses the cost of running your own proxy.
| Monthly requests | Direct cost | Gateway total | Overhead |
|---|
The convenience layer has a price that scales with you
An LLM gateway buys real things — one API across many providers, automatic fallback when a model is down, caching, logging and centralized spend limits — and charges for them as a percentage markup on tokens, a per-request fee, a flat monthly platform fee, or a mix. The trap is that a percentage markup is invisible when you are small and enormous when you are large: 5% on a $200 bill is a rounding error, but 5% on a $40,000 bill is a full-time engineer's salary a year, spent on a feature you could self-host with an open-source proxy. This calculator separates the raw provider cost from the gateway's cut so you can see the effective markup and the annual overhead in one place, then decide route by route. Compare the raw token prices on the model price comparison, recover repeated-prompt spend with prompt caching, and model the whole bill on the API stack cost calculator.
How this calculator works
The AI Gateway Markup Calculator computes the total monthly cost of routing your LLM traffic through a reseller gateway or router such as OpenRouter, Portkey, Helicone, or LiteLLM cloud, and separates that total into the underlying provider cost and the extra you pay the gateway on top. It starts from your monthly requests multiplied by the input and output tokens per request, prices those tokens at the provider's input and output rates per million, and then layers on the three ways gateways add cost: a percentage markup applied to the token spend, a per-request fee that scales with call volume, and a flat platform fee charged each month. The main drivers are your request volume and average token counts, since those set the base spend that every percentage and per-request charge is calculated against.
The key trade-off to watch is how each fee type behaves as you scale. A percentage markup grows in direct proportion to your token spend, so it becomes the dominant cost for high-token or high-volume workloads, while a per-request fee hurts most when you send many small calls with few tokens each. A flat platform fee is the opposite: it barely matters at high volume but can be the largest line item for a small project. A practical tip is to enter your realistic traffic and then compare the gateway's added cost against what you would pay the provider directly, since the convenience, routing, and observability features a gateway offers are only worth it if the markup stays small relative to your base spend.
Frequently asked questions
What does an AI gateway actually charge for?
An AI gateway or LLM router sits between your app and the model providers, giving you one API for many models, automatic fallback, caching, logging and spend controls. It pays for that by adding cost in one of three ways: a percentage markup on the underlying token price, a flat fee per request, or a monthly platform subscription — and some combine all three. A pure percentage markup scales directly with your token spend, so it stays cheap on light traffic but grows without limit as you scale. A per-request fee punishes high request counts with small payloads. A flat monthly fee is fixed regardless of volume, so it is expensive when you are small and negligible once you are large. This calculator adds up whichever apply and shows the real overhead.
Is a gateway markup worth paying?
It depends on what the overhead buys you and how big it is. On low volume, a few percent markup is trivial and the reliability, unified billing and instant model-switching are easily worth it. At scale the same percentage can become thousands of dollars a month, at which point self-hosting an open-source proxy such as LiteLLM and calling providers directly often pays for an engineer's time many times over. The decision hinges on the effective markup this tool reports: compare that annual figure against the engineering cost of running your own routing and caching.
How do I cut gateway overhead without losing the features?
Route only the traffic that needs the gateway's features through it and send high-volume, stable calls directly to the provider. Turn on caching so repeated prompts never reach the metered path. Watch for gateways that mark up cached tokens at the full rate. And once your percentage-based overhead crosses the cost of a self-hosted open-source proxy, migrate the heavy routes off the paid gateway while keeping it for the long tail of models you rarely call. The cumulative overhead table below shows exactly when that crossover is worth the migration effort.