Skip to content

Glossary

Token

The unit a model reads, writes, and bills in: in English roughly three quarters of a word, in Serbian noticeably less.

Text is cut into pieces that appear often — whole common words, parts of rarer ones, punctuation on its own. Our alphabet fares worse than English: the same sentence in Serbian costs more tokens than in English, and a price list per million tokens therefore does not mean the same thing in both languages.

Prices are quoted separately for input and output, and output is the expensive side, usually several times over. A cached input is cheaper still, which is why the shape of a prompt — what stays the same and what changes — decides the bill more than the model choice does.