Lesson 2 of 7
What's a token?
5 min read
Pricing pages talk about 'tokens' and dollars per million of them. What's a token — and why is that the unit you're billed in?
The unit of AI billing
A [token](glossary://token) is a chunk of text — roughly ¾ of a word. A rule of thumb: ~4 characters ≈ 1 token, and 100 tokens ≈ 75 words. Models read and write in tokens, so tokens are also what you pay in. Both your input and the model's output are counted this way.
Tokens, not words, are what a model reads, writes, and bills.
Why not just count words?
Tokens don't line up neatly with words: common words are one token, rare or long ones split into several, and spaces and punctuation count too. That's why the same idea costs more in some languages than others — and why 'price per million tokens' is the number on every pricing page.
A token isn't a word — pricing is per token, so word counts only estimate the bill.
To sanity-check a price, convert: multiply your word count by ~1.3 to get tokens, then apply the per-million rate.
Token basics
- —~4 characters ≈ 1 token; 100 tokens ≈ 75 words.
- —You're billed per token, for input and output.
- —Rare/long words and other languages use more tokens.
Roughly how much text is 100 tokens?
Continue in the app
Take the whole What AI Costs course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.