Models do not process letters or whole words. Before text reaches a model, a tokenizer converts it into a sequence of tokens drawn from a fixed vocabulary. Common words are often a single token. Rare or long words are split into several pieces. Numbers, code and non-English text tend to use more tokens for the same number of characters.
Tokens matter to data sellers because buyers size datasets in them. A count of rows says little when one row is a short reply and another is a long report. A token count shows how much text there really is. Each model family has its own tokenizer, so the same file yields somewhat different totals depending on which is used. Any count given to a buyer should name the tokenizer or be labeled an estimate. Count the content fields, not the surrounding JSON keys and punctuation, which repeat on every line. Reporting the distribution of tokens per record, as well as the total, tells the buyer whether records are uniformly sized or dominated by a few long ones.
The free counter on this site gives an estimate in the browser for pasted text or a file.