token-tax · measured, not estimated

Token premium per language

Every tokenizer, every language

Premium vs English on identical FLORES-200 content. Stronger contrast with the background = more expensive (see legend). Tokenizers sorted by geometric-mean premium (best first). Hover a cell.

What you pay, and how much fits

Input price per 1M English-equivalent words (the same content), and how many English-equivalent words of Telugu content fit in each model's context window. Prices from prices.yaml, checked .

Table view

Premium vs English for the selected tokenizer, with raw counts.