Tokens matter commercially because they are the meter. A workflow that re-sends the same brand book on every call is paying for it every time, which is why caching and retrieval are cost decisions as much as engineering ones.
They also bound what a model can hold at once. A context window measured in tokens is the real constraint on how much of your brand, your brief and your prior work can be in the room during a single decision.
QUESTIONS PEOPLE ASK
How are AI models priced?
Usually per token of input and output, at different rates, with discounts for cached input. Image and video models bill per generation or per second of output instead.