What is a token in an LLM?

A token is the small chunk of text an LLM actually reads and writes — usually a word, part of a word, or a piece of punctuation. Language models don't see letters or whole sentences directly; they break text into tokens first. As a rough rule of thumb in English, one token is about four characters, and 100 tokens is roughly 75 words.

The word cat might be one token, while unbelievable could split into un, believ, and able.

Why do tokens matter in practice?

Tokens are the real unit behind how you use and pay for an AI model. Three reasons stand out:

• Cost — most providers charge per token, for both what you send and what you get back, so a longer prompt or answer costs more.
• Limits — a model's context window, meaning how much it can consider at once, is measured in tokens, not words or pages.
• Speed — the model generates one token at a time, so a longer answer literally takes more steps and more time.

If you ever work with an AI API, tokens are the unit you're really budgeting around.

How is text split into tokens?

Before a model reads anything, a piece of software called a tokenizer chops the text into pieces from a fixed vocabulary it learned in advance. Common words often become a single token, while rarer or longer words get broken into several. Spaces, punctuation, and even emoji count too.

The same word can split differently depending on capitalization or a leading space, which is why token counts rarely match word counts exactly. Different models use different tokenizers, so the same sentence can cost a slightly different number of tokens on different systems.

Why does an LLM miscount the letters in a word?

A model sees tokens, not individual characters, so it has no clean view of spelling. Ask how many r's are in strawberry and it can stumble, because the word arrived as one or two tokens rather than a string of separate letters. The same blind spot explains why very long documents get cut off — once the text passes the token limit of the context window, the extra tokens simply don't fit, and the model never sees them.