Skip to main content
A token is the unit a model reads and writes. It is usually a piece of a word, not a whole word or a character. Fireworks bills Serverless usage in three token counts:
  • Prompt tokens are what you send.
  • Cached prompt tokens are the part of that prompt Fireworks already computed. They cost less. The discount depends on the model.
  • Generated tokens are what the model writes back.
The usage object on each response is the source of truth for those counts. Prices are on Serverless pricing. More words are in the glossary.