Tokens: the bites AI reads and writes

AI ovens don't read letter by letter or word by word. They read in small bites called tokens. Knowing about tokens helps you understand why some messages cost more than others, and why an oven can only keep so much in mind.

What is a token?

A token is a small piece of text. Short, common words like "cake" are usually one token. Longer or rarer words are cut into two or three pieces. Spaces and punctuation count too. As a rough guide, 1,000 tokens is about 750 English words, or about one and a half pages of a book.

Count them yourself

Type anything in the box below. Granny counts the tokens on this PC, the same way OpenAI's newest ovens do, and shows where each bite starts and ends. Nothing is sent anywhere.

Try a long word like "unbelievably", a number like "2026", or an emoji. Other languages often need more tokens for the same idea.

In and out

Every time you bake, two kinds of tokens are counted:

• Input tokens are what you send: your message, your baker's instructions, the chat so far and any document pieces.
• Output tokens are what the oven writes back. They usually cost several times more than input tokens.

Under every answer in GrannyBakes you'll see both numbers, for example "52 in / 140 out tokens", plus the cost.

How long can an answer be?

You can set an answer length limit (sometimes called "max output tokens"). The oven stops writing when it reaches it, even in the middle of a sentence. Thinking ovens count their hidden thinking against that limit too, so leave it empty unless you really need short answers.

Granny's rule of thumb: count words, then add a third: 300 words is roughly 400 tokens.