Token
The unit a model actually reads and writes — a sub-word chunk, not a character or a word.
A token is a piece of text produced by a tokenizer, typically a common sub-word (“play”, “‑ing”). Models process tokens, price by tokens, and are bounded by a token context window. This is why models miscount letters in a word: they never see the letters, only the token.
Worked example: “strawberry” may be a single token, so when asked how many r’s it contains, the model has never seen the individual letters — it is guessing about a thing it cannot perceive. In English, one token averages ~4 characters (~0.75 words). Gotcha: cost, context limits, and rate limits are all denominated in tokens, not words or characters — so “keep it under 500 words” is the wrong budget; count tokens, and remember code and non-English text pack fewer characters per token.