Credits
A credit is a thousandth of a US dollar by default: 1,000 credits are $1.00. An organization's members share one balance, held by its owner. A Gateway organization is billed on its own: its credits don't pay for a Workflow organization or a Search account, even one the same person holds. When an organization's balance is first opened it receives a one-time starter grant, $25.00 by default, which expires after 365 days.
The owner adds credits in Settings › Credits & billing, or turns on Auto-reload there, which adds a set amount when the balance drops below a threshold. The smallest automatic reload is $5.00.
What a call costs
Each answer a call is given reports the tokens it used, and each kind of token is priced at the rate the offering lists
under pricing in the catalog:
pricing field |
What it prices |
|---|---|
promptTokens |
Input tokens, per million. |
completionTokens |
Output tokens, per million. |
cacheReadMultiplier |
Input read from the provider's cache, as a multiple of promptTokens. |
cacheWriteMultiplier |
Input written to the provider's cache, as a multiple of promptTokens. |
thoughtTokens |
Reasoning tokens a provider counts apart from its output, per million. |
perUnit |
For a model priced by the unit, price is US dollars for one unit: a page an OCR model read, or a minute of the recording a transcription model transcribed, counted in fractions of a minute. |
A request's costs are added up and charged when the request ends, rounded up to a whole credit. Work done for the call is part of it: a model that converts a file is charged as that model's provider charges it, by its tokens, the pages it read or the minutes of audio it transcribed.
A call that runs on your own provider key, sent with the call or saved for your organization, is billed by that provider to your account, and charged from your credits a share of the price the catalog lists: 5% by default, a setting of the deployment. The same share applies to Search and to workflow runs on your saved keys. See bring your own keys.
Seeing what you spent
Every attempt on a call's record carries costUsd, what it cost at the price of the offering it ran on. Read one call
with Get a generation, or list your organization's calls and their totals by day
as described in usage and costs.
Choosing on price
Gateway orders the offerings that can serve a call by price, cheapest first, unless you rank them yourself. Offerings on your organization's prioritized keys go ahead of that order. See choosing a model.