Token
Also called: tokens
In one sentence
The unit of text a model actually processes — usually a word, part of a word, or punctuation.
In more depth
Tokenisers are built by finding character sequences that occur often enough in text to deserve their own symbol. Common words become a single token; rarer words split into several.
Tokens are the unit of both billing and context limits, and different model families use different tokenisers — so a token count is only meaningful alongside the tokeniser that produced it.