What one call actually costs
Pick a call size and see the price on every model, in sats and in your currency. Everything on this page is computed from the live broker's /v1/pricebook with the same formula the signed receipts use, so you can check any number here against a real receipt.
Call size
| Model | Provider list price USD per 1M in / out |
Metered call (channel) |
With reasoning (estimate) |
x402 quote (live, per call) |
|---|---|---|---|---|
| Loading the live price book… | ||||
How to read the columns
Metered call. What a prepaid channel is charged for exactly this many tokens: ceil((in × inRate + out × outRate) / 1000) sats, at the per-1k rates in the price book. This is the frozen formula every signed usage receipt carries, so the receipt for a real call of this size shows this same number.
With reasoning. Thinking models also bill reasoning tokens you never see. This column adds a typical amount (the broker's published overhead of 2.5× the visible output), capped by the output allowance, since reasoning and the answer share max_tokens. It assumes max_tokens is set to the output size (thinking models get at least the floor below). It is an estimate: the receipt for a real call lists the actual hidden tokens as signed, separate fields, never more than your output limit.
x402 quote. Pay-per-call has no channel, so you pay the quote up front. The quote reserves for your max_tokens, so this column is a real, unpaid 402 quote fetched just now with max_tokens set to the output size above. Thinking models reserve at least 1024 output tokens. Set max_tokens close to what you need to keep the quote small.
Web search is extra and priced per search: see webSearch in /v1/pricebook. Volume accounts get discounts; the discount is a signed field on each receipt.