Billing rules

Three billing factors: model unit price × group ratio × usage. Ratios come from the pricing API; usage logs are the ground truth.

Billing modes

Usage-based: input / output tokens priced separately — output is usually more expensive (see the completion ratio in the marketplace).
Per-call: some models charge a fixed price per call regardless of tokens.
Group ratio: final price = base price × group_ratio (table below).
Pre-hold & settle: streaming requests hold quota first and settle to actual usage afterwards — brief balance jumps in the logs are normal.

Syncing live data…