Explain / Token

What is a token?

The pieces a model reads and writes in — roughly a short word each. Limits and prices are counted in tokens, not pages. Nothing to do with a login token.

In plain English

A token is the piece a model actually reads and writes text in. Not quite a word, not quite a letter, roughly a short word or a chunk of a longer one. “Unbelievable” might get split into three separate pieces, for instance.

This matters because every limit you’ll hit, pricing, context windows, all of it, is counted in tokens rather than words or pages. So when someone asks “how long can my document be”, the honest answer is always in tokens, because that’s the actual unit the system is working in underneath.

A worked example

A rough rule of thumb: one page of ordinary English works out to a few hundred tokens. So a two-page memo is a few hundred tokens, and a 200-page report is tens of thousands. The window sizes assistants quote are in the same unit, which is why the two numbers are worth putting side by side before you upload.

That’s why “can it read my whole report” doesn’t have a yes-or-no answer without knowing the length. Some assistants take a long document in one go, some don’t, and the answer loops straight back to the context window and the same fix: break it into chunks.

The usual mix-up

People hear “token” and think of an API token or a login token, the password-like key you use to access a system. Same word, completely different thing. One is a security credential, the other is just a unit of text. There’s no relationship between them at all, it’s just an unfortunate bit of shared vocabulary that trips people up constantly.

Where you’ll use it

Also comes up in: automate repetitive tasks, update itself.

All the words