Pricing and cost calculation
Understand input, output, cache, and per-request billing using actual line prices.
Find prices
Open the model in the catalog and inspect the selected line, billing unit, and input/output prices. Token prices use USD / 1M tokens. Image, video, and audio models may use per-request or other explicitly listed units.
Recheck prices before launch or line changes. Another provider's prices, a different line, or an old screenshot is not the settlement basis for the current request.
Estimate token costs
When there are no cache or other billing categories:
estimated_cost_usd = input_tokens / 1_000_000 * input_price_usd_per_1m
+ output_tokens / 1_000_000 * output_price_usd_per_1mPrices are in USD / 1M tokens. For cache reads, writes, or other categories, split usage into non-overlapping billable components. Do not count cached tokens again as ordinary input when they are already included in the input total.
An output limit is a request parameter, not guaranteed output volume. Tool loops, conversation history, client helper requests, and retries may increase workflow usage. A single-request estimate is not a whole-workflow budget.
Other billing units
Estimate using the units and parameters shown on the model page. Resolution, quality, count, or duration may affect price. Do not apply ordinary text token prices to image or video requests.
Reconcile
Estimates support budgeting; inspect usage for settlement. Match the account, key, model, and time range, then check caches, helper requests, and asynchronous jobs. Investigate timeouts, cancellations, and failures against actual records rather than inferring charges from a client error alone.