Skip to main content
You pay per token. There is no charge per request, and failed requests are free. Prices are in US dollars per million tokens.

​
How tokens are counted

Every response reports its tokens in usage:
  • Input is prompt_tokens minus cached_tokens.
  • Cached input is cached_tokens: input that was served from the prompt cache. See Prompt caching.
  • Output is completion_tokens, including the model’s thinking when reasoning is on.
The request above costs (176 × $0.50 + 1,024 × $0.10 + 300 × $0.90) ÷ 1,000,000 = $0.00046. Streamed responses count the same way. fastrun reads the token counts from the end of the stream even when you don’t ask for them with stream_options.include_usage. Only successful responses (HTTP 2xx) are counted. If you close a stream early, generation stops.

​
Your usage

The Usage page of the dashboard shows requests, tokens and the estimated cost, by UTC day and by key, for the last 30 days. A request appears there as soon as its response has ended.