AI 'tokens', explained in plain English

Spend five minutes reading about AI and you hit the word tokens. Prices are “per million tokens.” Models have a “token limit.” Something got cut off because it “ran out of tokens.” So what is a token, really, and why does everything come back to it?

In 60 seconds

A token is a small chunk of text. Not quite a letter, not quite a word, usually a short word or a piece of a longer one. AI language models don’t read text the way we do; they chop it into tokens and work with those. Every prompt you send and every reply you get is measured in tokens. That’s why tokens show up in three places at once: what you pay, how much the model can handle at a time, and how long its answers can be.

The plain-English version

Imagine the model reads with a very particular pair of scissors. It doesn’t cut on spaces, and it doesn’t cut on letters, it cuts text into common little pieces. A simple word like “dog” is one snip. A longer or unusual word like “tokenization” might get cut into three or four pieces. As a rough rule of thumb in English, one token is about four characters, and 100 words is around 130-150 tokens.

Everything else follows from that:

• Billing: models charge per token, in and out. A long document you paste in costs tokens; a long reply costs tokens too.
• The context window: a model can only hold so many tokens in mind at once, that limit is the “context window.” Go over it and the oldest stuff falls off the edge, which is why a very long chat can start “forgetting” the beginning.
• Length limits: when a reply stops mid-sentence, it often just hit a token ceiling.

Why it matters now

As people feed AI whole books, codebases, and long chat histories, tokens became the thing you actually budget. Bigger context windows are a headline feature precisely because they mean more tokens in mind at once, more document, more memory, more conversation before things fall off the edge. If you use AI tools seriously, “how many tokens does this cost / fit?” is the question hiding behind most of the others.

What people get wrong

• “One token = one word.” Close, but no. Short common words are one token; longer or rare words split into several. Spaces and punctuation count too.
• “Tokens are the same across models.” They aren’t. Different models chop text differently, so the same sentence can be a different number of tokens depending on the model.
• “More context window is always better.” More headroom helps, but stuffing a huge amount in costs more and can actually dilute the model’s focus. Room to hold text isn’t the same as using it well.

The one-line takeaway

A token is the bite-sized chunk of text that AI models actually read and write in, and once you picture text as tokens, the pricing, the limits, and the “why did it forget?” moments all stop being mysterious.