What is a token?
A token is the small unit of text a large language model reads and produces, often a word or part of a word. AI providers count tokens to set pricing, speed limits and how much material a model can handle at once.
Models do not see words the way people do. They break text into tokens, which might be a whole short word, a piece of a longer one, a number or a punctuation mark. As a rough rule of thumb, a page of ordinary business writing runs to several hundred tokens, though the exact count depends on the model and the language.
Tokens matter for two reasons. First, API pricing is usually quoted per million tokens in and out, so a workflow that sends whole documents on every request costs more than one that sends only the relevant section. Second, every model has a limit on how many tokens it can consider at once, known as the context window.
For most owners, tokens only become a real line item when you move from a per-seat chat subscription to an automation that calls a model many times a day. At that point, estimate volume before you build: requests per day, typical length, and the provider's rates. Check current pricing directly, since it changes often.
Example
For example, a staffing agency that wants AI to summarize every incoming resume would estimate tokens first. If it receives a few hundred resumes a week, the team multiplies typical resume length by volume, adds the summary length, and compares the monthly cost against recruiter hours saved before committing to a build. The numbers here are illustrative.