Billing rules

Unified token rate

Current token billing adds input, output, cache-read, and cache-write tokens and applies one credits_per_unit rate per 1,000 tokens. Public pages show “X Credits per 1 million tokens (same rate for input and output)” without separate input/output prices. Refer to the current public models and pricing pages for X.

The server rounds raw Credit charges upward to eight decimal places. Do not recalculate charges in clients; use the decimal strings returned in responses, usage, and billing details.

Reservations and failures

A request may show a temporary reservation. At completion, actual usage is settled and the remainder released. Interrupted streams settle usage already generated. A failure before billable usage does not fabricate successful usage; treat usage and transaction records as authoritative.

Personal statement periods are UTC calendar months. Pages may display dates in the account timezone; these are separate concepts.