Price AI credits without guessing

It turns four assumptions into a price: the provider's input price, its output price, the token mix of a request and the fixed cost per call. It then applies your target gross margin and divides the result by the value you assign to one credit.

The maths is intentionally small and deterministic. It does not call a model, fetch a price list or read your billing provider. The result is an estimate for planning, not a quote for a customer.

Take provider prices from your contract, not from a memory of a launch-week promotion. Take token counts from real request logs or a measured sample, including the system prompt and any retrieved context. Set the credit value from the plan you intend to sell, and choose a margin that still leaves room for payment fees and refunds.

Run a low, expected and high token case. If the price changes too much, your credit unit is too small or your margin assumption is too tight. A single average hides the long requests that create the bill.

This is gross margin. Taxes, Stripe or other payment fees, refunds, failed provider calls, retries, storage, observability and free usage are outside the calculation. A provider's free tier can make the first month look better than the steady state.

The calculator also does not model rate limits or abuse. A per-key limit and a credit reserve protect spend at runtime; this page only helps set the numbers before those controls are exercised.

A credit is whatever your product says it is. The calculator converts a target price into that unit; it does not claim that one token equals one credit or that a credit has an intrinsic value. Publish the conversion and round the quantity in a way customers can understand.

Keep the provider cost, the credit ledger and the payment amount as separate records. If a refund happens, the ledger explains what was used; it should not silently rewrite the price of a past request.