Tokens
The chunks a model reads and writes in — roughly ¾ of a word each. Everything is priced and measured in them.
Think of it like
Syllables for a machine — not quite letters, not quite whole words.
Example
“unbelievable” might be three tokens: un-believ-able. Your bill counts every single one.
How it actually works
Tokenization is why models miscount the letters in a word or fumble rare strings — they never saw the letters, only the chunks. It also explains why some languages cost more: they tokenize less efficiently, so the same sentence spends more tokens.
For product teams
The billing and length unit for AI — costs and limits are counted in these, not words.
For engineers
Sub-word units from the tokenizer; model I/O and pricing are both measured per token.
Related
- Context window — The unit the context window is measured in.
- Embeddings — Cousin of embeddings — both turn text into numbers.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome