Tokenizer tax makes AI costlier for less widely spoken languages
A study finds some languages require far more tokens than others, making artificial intelligence more expensive for the people who speak them.
By GennnAi··1 min read
Published by the Coalition for Language Equity in AI Research, the study found that among the 20 most common languages, the costliest uses 1.5 to 5 times more tokens than the cheapest. With 200 languages, the gap reached 17 times in one model. Script matters: Burmese, Tibetan and Khmer are most penalized. Latency and usable context also suffer. One investor asks labs for language-by-language scorecards.
Source: Rest of World, published on October 26, 2026.