Tokens

Models do not see characters the way we do. Text is broken into tokens: pieces of words, punctuation, sometimes whole short words. “Token” is also how providers meter usage and how context windows are sized.

A shorter prompt is not only nicer to read. It can be cheaper and leave more room for the answer and the documents you stuffed in. That is why people joke about caveman mode and token poker.

When a slide says “128k context,” it is talking about tokens, not pages of a paperback. The conversion is rough and language-dependent; treat marketing numbers as order-of-magnitude, not exact page counts.