Compare three ways to price an AI product on your real customer mix. See revenue and margin for each model, and which customers you lose money on.
Set a price for each model, then see how they perform.
Group customers by how much they use.
Monthly revenue by pricing model.
| Segment | Your cost | Flat | Per token | Credits |
|---|
Each cell shows monthly revenue with the gross margin below it.
| Flat | Per token | Credits | |
|---|---|---|---|
| Bill predictable for customer | High | Low | Medium |
| Revenue follows your cost | Low | High | High |
| Easy for buyers to understand | High | Low | Medium |
| Protects margin on heavy users | Low | High | High |
A model, not a forecast. It assumes the same usage under every pricing model and ignores churn, discounts and taxes.
Flat subscriptions are simple to sell but only profitable for the average user. In AI products a small group of heavy users can consume most of the inference budget. Pay-per-token pricing tracks your cost exactly but is hard for buyers to predict. Credits sit in the middle: a unit tied to value, prepaid for predictability, with usage-based economics underneath.
Our guide AI credits vs tokens covers the trade-offs in depth, and the glossary explains credit-based pricing. To design tiers with included usage and overage, use the Usage Pricing Plan Builder.
Breakage is the share of prepaid credits customers buy but never use. It is extra margin for you, but treat it carefully: relying on high breakage can hurt trust, and accounting rules may require you to recognise it over time.
Start from cost: credits per request × price per credit should comfortably exceed your cost per request. Then round to numbers that are easy to explain, such as 1 credit for a simple action and 5 for a heavy one.
A flat fee is fixed while cost grows with every request. Beyond the break-even usage, a customer costs more than they pay and other customers subsidise them.
No. It runs in your browser, and the share link keeps your inputs in the URL fragment, which is not sent to our servers.