Usage and billing

You pay per token at the rates shown in GET /v1/models. Charges go to the organization that owns the API key. Paid models need a positive credit balance — if credits run out, new requests are rejected until you top up.

What a request costs

Prompt

Fresh prompt tokens at the prompt rate. When the model lists a cached-prompt price, repeated prompt tokens bill at that lower rate. prompt_tokens in the usage object includes cached tokens.

Completion

Completion tokens at the completion rate. Reasoning tokens are included in completion_tokens; when the model lists a separate reasoning price, they bill at that rate instead.

Multiple completions

n from 1 to 8 multiplies completion cost. Each choice is billed.

Cancelled stream

Prompt plus whatever completion tokens were already sent. Closing the connection does not make the request free.

usage.cost

USD billed for this request. Present on JSON completions and on the stream usage frame.

A request that fails before any work starts — for example a 400 or 401 — costs nothing.

Top-up

Admins add credits at Billing. Packs are $5, $10, $25, $50, $100, $250, and $500. Checkout is handled by Dodo. After a successful payment the page returns with topup=success.

Auto top-up is optional. Admins authorize a card, then set a credit threshold and a recharge amount. When the balance drops below the threshold, the card is charged. Some cards take up to 48 hours. Members can view billing. Only admins change auto top-up or open the customer portal for cards and tax invoices.

Usage in the dashboard

Home shows token totals, model mix, and credit balance. Filter by date range and by API key versus playground. You can download a usage report from that page.

Rate limits

Each organization defaults to 60 requests per minute. Over the limit returns 429 with code rate_limited and a Retry-After header. Exhausted credits return the same status with code insufficient_credits. See Errors. Email support if you need a higher RPM.