Billing guide
Prepay and settle against actual usage.
Precharge and settlement
Requests check balance and may reserve credit first. The final charge is settled against actual usage, returning or collecting the difference. Check final usage logs and wallet records rather than treating a temporary reservation as the charge.
Tokens and caching
Input, output, cache reads and cache writes may have separate rates. Cache pricing applies only where the selected model and billing configuration support it and actual usage reports it. Do not assume every input token is cached.
See Models and pricing for rates. Account groups may affect the final rate; check the model and group before calling.
Introductory and volume benefits
Published list prices remain in effect. During the launch enrollment window, eligible first-time top-ups receive 90% list-price billing for 14 days from the first paid request, capped at $100 of list-price usage and 100 customers. Closing enrollment preserves previously issued offers. Check the pricing page for campaign status and the wallet for personal eligibility, expiry and remaining allowance.
The previous calendar month's net paid API usage in UTC+8 sets the following month's tier across all keys in an account. Below $100, list prices apply; at $100, pay 95%; at $500, pay 92%; at $2,000, standard models pay 90% and flagship models pay 85%. Flagship models are Fable 5, Fable 5.1, Opus 5.5 and GPT-6 Astra.
Input, output, cache reads and cache writes are included; tool and other additional charges are billed separately. The best introductory or volume rate applies without multiplication. Requests crossing the introductory cap split eligible and ordinary portions. Benefits apply to standard-group wallet billing; special-price groups and subscriptions do not stack with this program.
Top-ups and gifted credit are not consumption. Net paid usage counts settled consumption of cash-funded balance. Refunds reduce the month in which they complete, do not rewrite an already granted tier, and do not replenish introductory allowance. A request retains its starting price version and applicable calendar month; consumption is assigned to the month of settlement. Usage logs record list-price, savings and actual charge amounts.
Two types of refusal
A 403 content_policy_refusal returns the reserved credit and is not charged. A successful response containing a refusal is processed usage and is billed normally. These are different cases.
Records and questions
Use logs to check tokens, charge and local request ID. Payment availability, balance expiry, refunds and invoices follow formal information shown before payment. Contact support if the rules are unpublished or unclear.
Still have a question?
Support details have not been published.