A token is the smallest unit of text a language model works with: it might be a whole word, part of a word, or a single punctuation mark. Before the model reads your text, it splits that text into a string of tokens and processes those. Grasp this, and you can make sense of why usage limits and costs are counted the way they are.
A One-Sentence Definition
A token is a chunk of text the model treats as one unit. You hand it a sentence, and it applies a fixed set of rules to break that sentence into tokens, each mapped to a number the model actually computes on. Common words are often a single token, while rare or long words get split into several. Different languages split differently, so the same idea can take more tokens in one language than another.
A Concrete Example
Take the word “cat”: most models treat it as a single token, while a long word like “internationalization” gets broken into several. A short phrase splits into a handful of tokens, and the exact count depends on which model’s tokenizer you are using. You do not need to count them by hand; the useful rule is simply that more text means more tokens.
How It Differs From Related Ideas
A token is not the same as a “word” or a “character,” even though people use those terms loosely. One English word can be a single token or split into two or three. A character count is something you can tally yourself, whereas a token count is decided by the model’s own splitting rules, so you can only estimate it with a tool. When a model advertises a limit “in tokens,” do not read that as the same number of words.
When It Actually Matters
Tokens stop being abstract the moment you care about cost, how much text you can fit in, or response speed. Usage and limits are usually counted in tokens, and long documents or long conversations eat through more of them. You can feel this directly in ChatGPT or Claude: paste in a long document and ask for a summary, and it processes far more tokens than a one-line question does. In ChatGPT, which models and how much usage you get vary by plan and region; in Claude, usage allowances and advanced features depend on the plan too. So whether the same long text fits within your account’s limits can differ from one setup to the next.
