Billing & caching

Metering#

Four token classes: input, output, cache read, cache write. Rates per model are on the catalog. Before the model runs, the gateway holds the worst case against your balance; after, it settles to what actually ran. Money is integer nanodollars end to end.

The receipt#

Every response reports what you asked for, what served it, why, and the settled cost, in the receipt headers. The same family carries the per-request cost ceiling.

Prompt caching#

Cache reads bill at a fraction of the input rate and cache writes at a premium over it; the exact rates per model are on the catalog. Which wire carries cache directives and how breakpoints are sent is in the API reference.

When the money runs out#

{"error": {"type": "insufficient_allowance", "docs_url": "https://conifer.build/account", ...}}

The balance can’t cover the request’s worst case, or billing isn’t set up. No model was called and nothing was charged. Fix it on /account.