Inference Cost
Also called Token Cost, Cost per Token, and AI Tokens
What it costs to run an AI model each time it answers, usually billed by the token, a chunk of text of roughly three-quarters of an English word, counted on both what you send and what comes back.
Why it matters
Every use of an AI feature costs money, unlike most software, where serving one more user is nearly free. Long prompts, large models and agents that loop many times multiply the bill, so inference cost sits directly in your cost of goods sold and decides whether a low-priced plan can make money.
For example
A startup finds that its heaviest users each trigger thousands of model calls a month, costing more than their subscription, so it moves routine requests to a smaller, cheaper model.
Go deeper
Related terms
Also in Starting Up
See how founders actually handle this on Money, pricing and metrics, part of the Starting Up hub.
eChai Partner Brands
The partnership →
