What is a token in AI?

A token is a small piece of text, often a short word or part of a longer word, that AI models read and write. AI models break all text into tokens before working with it. In English, 100 tokens is roughly 75 words, and a page of text is around 650 tokens. AI providers count both usage limits and prices in tokens.

Why it matters for a business

Tokens are the unit of both the limit and the bill. The context window, how much a model can take in at once, is measured in tokens, and model providers charge per token on the way in and on the way out. Most of the tokens in a business request are text nobody typed: standing instructions, a pasted document, the records a lookup pulled in. So what a workflow costs depends far more on how much text each run carries than on the price per token.

Example

Let’s say someone pastes a 12-page contract into an assistant and asks for a short summary. About 8,000 tokens go in and about 500 come back, and both counts appear on the bill.

Go deeper

What a context window is, and what happens when it fills up

Related terms

• Token pricing
• Context window
• Inference
• Large language model (LLM)

All terms, A to Z