Pricing and billing

Check RMB prices, billing units and your key's group before calling a model.

Use live prices#

Customer prices and payments are denominated in CNY. Groups, context tiers, request parameters and time windows can affect prices. This guide does not freeze model rates: check the catalog and the applicable prices recorded for the request.

Text and cache usage#

Text charges commonly include input, output and applicable cache categories. CNY / 1M Token means the RMB price per million tokens, not per call. Tokens are not a count of characters.

Cache hits depend on actual usage and model rules. Cache-read rates may appear only in expanded details; a card showing input and output does not imply cache pricing is absent. Separately priced cache tokens are not billed again as ordinary input.

Images and video#

Images may be billed by image, request or tokens; video by task, second or reported usage. A per-second price is not a per-video price. Duration, size, quality, resolution and reference video may affect cost.

Groups and time windows#

Select the same group in the pricing view as on your key. Groups can affect eligible channels and charges.

Peak/off-peak rules apply only when configured for that model. Asia/Shanghai means Beijing time. Monday–Friday conditions do not automatically account for Chinese public holidays or adjusted working days. Boundaries and multipliers follow the model's rules.

Reservations and settlement#

The platform may reserve estimated quota and later reconcile it against actual usage or a task result. An in-progress reservation is not necessarily the final charge. Missing upstream usage may retain an estimate or reservation; inspect the request record.

A disconnect, cancellation or missing visible output does not prove the upstream did no work and does not guarantee zero charge. Duplicate submissions may create additional charges.

Check a charge#

In usage records, inspect model, group, input/output/cache counts, pricing tier, time conditions and final amount. For video, also inspect task status. Give support a request ID, time and redacted screenshot, never your API key.