A token is the unit a language model actually processes. Text gets split into tokens before the model reads it, and a token is usually a common word or a fragment of a longer one. As a rough guide, a token averages about four characters of English, so a page of text runs to several hundred tokens. Punctuation and spaces count too.
Tokens matter to founders for one blunt reason: they are the meter. API providers charge per token, counting both what you send and what the model sends back. A verbose system prompt repeated on every call, or an answer allowed to ramble, quietly inflates the bill in a way that is easy to miss.
Tokens also define the context window, the maximum amount of text a model can consider at once. Everything you supply plus everything it generates has to fit inside that budget. Exceed it and older content gets dropped or the request fails, which is why long documents often need to be chunked and retrieved rather than pasted whole.
A couple of habits pay off. Estimate token counts before shipping a feature, and remember that non-English text and code often use more tokens per word than plain English. Trimming boilerplate from prompts is one of the simplest ways to control both cost and latency, and small savings compound across millions of calls.