04 · How Models Run
Token
The unit a model reads & writes
A unit a model processes, often a word piece, character group, byte sequence, or other chunk chosen by its tokenizer. Word-to-token ratios vary by language, code, and tokenizer, so “about 750 English words per 1,000 tokens” is only a rough estimate.
Concrete example
A 500-word email is about 650 tokens — the unit you're billed in when you call an API.