Glossary term

Cost per token

Total model-related cost divided by the number of tokens used or processed.

The most commonly cited AI unit cost, and the easiest one to get wrong: it depends entirely on which cost is in the numerator. A vendor's own rate card, a negotiated enterprise discount, and a prepaid committed-use rate all produce different "cost per token" figures for the identical workload, which is exactly why this site's comparisons ask where the number comes from before repeating it.