Token
A token is the unit a language model reads and writes: roughly a fragment of a word, and the unit almost all AI usage is measured and billed in.
Models do not work in characters or words. Text is split into tokens, where common words are usually one and rarer ones break into several. As a rule of thumb English runs to about three-quarters of a word per token, which is why a document’s token count never quite matches its word count.
Tokens matter in practice for two reasons: the context window is a token budget, and cost is charged per token in and out. Anything that puts more material in front of the model — a longer conversation, more retrieved passages, a bigger system prompt — spends both. Retrieving well is therefore a cost decision as much as a quality one.