A token is the chunk of text a model reads at once. Not a word, not a letter: roughly four characters of English, so a short word might be one token and a long word might be three. Before the model sees your text at all, everything you send is chopped into these chunks, and everything it writes back comes out the same way.
Tokens matter because they are the unit the whole economy of AI runs on. Pricing is counted in tokens, in and out. Speed is counted in tokens. The limits on how much a model can read at once are counted in tokens. When a vendor quotes a price or a limit, tokens are the currency behind the number, whether the sales page says the word or not.
For an operator, this is the cost model, and it repays five minutes of attention. Long inputs cost more than short ones. Long answers cost more than tight ones. A workflow that pastes an entire document into every request pays for that whole document every single time, which is why summarizing once and reusing the summary, or fetching only the relevant pages, shows up in every serious AI build.
A concrete example. Two teams build the same contract review assistant. One sends the full contract with every question. The other sends only the clauses that matter for the question being asked. Same model, same answers, very different bills at the end of the month. Nothing about that difference is technical magic. It is just token arithmetic, and it is the kind of arithmetic worth doing before a tool goes live.
“Long documents cost more because everything is billed in tokens.”
// this page is the full text. the packaged PDF, all fifteen terms, comes with the free library.