Usage and API pricing
API use is billed separately from the included chat allowance. The configured retail rates are in US dollars per million tokens:
| Category | Price |
|---|---|
| Input | $0.50 |
| Output | $1.00 |
| Cached input, when authoritatively reported | $0.10 |
Cached input replaces the regular input rate for those tokens; it is not an additional charge on top of them. For example, 1,000 uncached input tokens and 500 output tokens cost $0.001 at these rates, before any separately disclosed applicable charges or taxes.
Read a usage receipt
Section titled “Read a usage receipt”A chat completion’s usage object uses prompt_tokens, completion_tokens,
and total_tokens. Cached input, when supported and reported, is a subset of
prompt tokens. Missing cache details do not prove a cache hit or a zero count.
An agent may make more than one model call while answering one request. The intended chargeable usage aggregates actual newly executed customer-agent calls, including planning and the final answer. Previous requests’ usage is not charged again; history sent as input to a new provider call is new input usage. Library compression accounting is separate and is not silently added to this scope.
Preview status
Section titled “Preview status”Authoritative usage collection and gateway settlement are prepared but have not completed combined production acceptance. Paid API activation remains off. The final contract must preserve actual counts, explicit cache support and failure settlement; token estimates are not an authoritative receipt.
Once enabled, use the API console’s Usage logs to inspect request activity and wallet deductions. Do not infer a successful charge from HTTP 200 alone: some safe availability responses can use that status. The planned policy is no customer charge for a trusted unavailable request, while retaining actual provider spend in internal accounting.