16. Billing and Quota#

• Billing model: you are billed on actual token usage. Chat / embedding endpoints settle by input + output token counts; images, audio, video, and others settle according to each model's own pricing rules.
• Usage fields: the usage object in chat responses (prompt_tokens / completion_tokens / total_tokens) reflects what that request consumed.
• Quota limit: each API Key can be given a quota; once cumulative usage reaches the limit, requests return 402.
• Balance: requests are rejected when the account balance is insufficient — top up in the console in time.
• Asynchronous tasks: submission pre-freezes an estimated quota; the task is settled on actual usage at completion, and canceling a task unfreezes and refunds the remainder.

The exact unit price of each model is as displayed in the console / model list.