Why charge by Token

Token-based charging reflects the actual computational resources consumed when processing text inputs and generating outputs. This method directly aligns costs with usage measured in fundamental text units.

Language models process text as token sequences, where each token represents a meaningful fragment (e.g., part of a word). Processing tokens demands significant computational power and memory. Charging per token ensures billing scales directly with model workload and processing complexity. This method applies universally across various input and output tasks like generation or analysis. It inherently promotes efficient input design while preventing arbitrary limitations common in per-request pricing.

This approach provides transparent and precise cost measurement. Users pay only for the precise resources their specific requests consume, regardless of input complexity or length. It allows providers to sustainably scale operations while offering users predictable costs tied directly to their unique workload intensity.