Pricing and Billing

The platform uses pay-as-you-go credits. Current public prices and any applicable account-specific price are shown on each model page.

How charges are calculated

A request can contain one or more billable meters. Text and embedding models commonly bill token usage, while image and video models can also bill by output count, size, quality, duration, or other model-specific attributes.

Each request is charged as follows:

1. Determine billing items: Billing items are selected from the request's actual usage and attributes. Different items are added together; when an item has several price tiers, only one matching tier applies. Items with zero usage are not charged.
2. Calculate each item exactly: Item charge = usage × unit price ÷ pricing unit (for prices quoted per million tokens, the pricing unit is 1,000,000). If an exclusive discount applies, the charge is then multiplied by the discount multiplier. Nothing is rounded at this step.
3. Round once per request: The item charges are added up, and the total is rounded up to the nearest 0.000001 credits. That is the final charge for the request. Rounding happens only once, at the end, so it adds less than 0.000001 credits.

The final amount is recorded in request logs.

Interrupted and failed requests

A request with no billable usage is not charged. If a stream is interrupted after usage has already been produced, that actual usage can still be charged. Retries or failover attempts that produce billable output can contribute to the final usage.

Use the model page for current prices and request logs for the authoritative amount of a completed call.