Pricing

API limits vary by model and usage tier, and are shared across all API keys in your personal account or organization. They measure requests per minute and, where applicable, tokens per minute. Token limits count uncached input plus output tokens.

Your tier increases automatically as your lifetime paid API usage invoices and credit purchases reach each tier threshold. This is based on total payments, not your remaining balance; promotional credits do not count. View your tier, current usage, and per-model limits in the dashboard’s Limits tab, or contact us for higher or custom limits.

Chat has separate hourly input and output token allowances, shown in the Chat plan. These reset at the start of the next clock hour.

Spending limits and per-key caps are soft. A request is admitted based on usage that has already been recorded, and a request only produces a usage record once its response completes. Requests that are in flight when the limit is reached finish and are billed, and new requests begin failing with insufficient_quota a few seconds after the crossing request completes.

An interrupted stream is billed for the tokens generated up to the interruption; requests the model rejects or that fail before a response starts are not billed. If you need a hard ceiling, bound your concurrency and per-request output length. See how usage is recorded.