Billing
Balance, held and available — kept as integers.
Settled on actual usage
Input, cache-hit input and output are each priced at the model's rate. The pricing page shows settlement rates per million tokens.
Holds and settlement
Before a request starts, balance is checked and a hold is placed; after completion it settles on the upstream-reported usage and releases the rest. Requests that never entered generation release the hold; interrupted requests with incomplete usage go to review and are settled or refunded after verification. Cache writes bill at the regular input rate; cache hits bill at the cached input rate.
Balance, held and available
Balance minus anything held by requests in flight is what a key can actually spend right now. Balance is what you have paid in, less what has settled. Held in flight is money reserved by requests that have not finished. Available is the difference, and it is what the next request is checked against.
Money is integers
Everything is stored in nano-dollars as integers. Floating point on a per-token price drifts, and on a percentage margin the drift shows up as whole percentage points, so it is integers end to end and rounding happens once, at settlement.
Daily token quotas
Each key can have its own daily token cap, accumulated per UTC day. When the cap is reached requests return 429; insufficient balance returns 402.
Balance & top-up
Check balance, available channels and promotions in Balance & top-up. After creating an order, pay the exact amount to the network and address shown; stop paying after expiry. Credit status is in top-up orders; call costs are in usage records.