Concepts
Billing
Multi-dimensional billing: token / cache / reasoning
The platform bills by model × dimension; different capabilities use different units. Each call's actual usage is recorded in the usage log and invoice.
Billing dimensions
| Dimension | Applies to | Notes |
|---|---|---|
input_token / output_token | chat / embedding / reasoning | text tokens, most common |
cache_read | chat | cache-hit input, lower unit price |
cache_write_5m / cache_write_1h | chat | explicit cache writes |
thinking / reasoning | reasoning models | chain-of-thought tokens |
Price structure
Each model's list price includes input_price / output_price (per million units, USD); billing_unit is currently token. Formula:
cost = Σ (unit usage × dimension unit price)- Text chat:
input_tokens × input_price + output_tokens × output_price(per million) - Cache hit:
cached_tokens × cache_read_price(lower unit price) - Reasoning:
reasoning_tokens × output_price
When you're charged
- Pre-deduct: on request entry, an estimated input amount is pre-deducted (
input_cost) to ensure sufficient balance. - Settle: after the response returns, the real
usage(input + output) is settled — refunding any difference.
Usage details are in the console's "Usage / Billing" view, drillable by model, dimension, and time. Cache-hit portions are billed at the lower cache-read unit price.