Aarva podcast

‘Tokeniser tax’: Why AI is more expensive for non-English users

18.11.2026
0:00
5:07
Spol 15 sekunder tilbage
Spol 15 sekunder frem
Why does it cost more to talk to a machine if you don't speak English?

Building artificial intelligence requires breaking human language down into tiny digital pieces called tokens. But as Ananya Bhattacharya points out in a 16 November 2026 report for Rest of World, those systems were built by and for English speakers. When someone types a prompt in a language with a non-Latin alphabet, like Burmese or Tibetan, the machine suddenly has to work much harder. It burns through more tokens, slowing down the response and driving up the price. It is a technical quirk that quietly decides who gets to afford the internet's next era.

An explanation of the "tokenizer tax" that makes artificial intelligence more expensive and slower for non-English users. Drawing on recent research, it shows how AI models process text using units called tokens, and why non-Latin scripts require far more of them to handle the same information. Details the financial and technical penalties this creates for developers outside the West, and outlines what AI labs must do to close the gap.

Read at source: Scroll.in

For more features and details, visit aarva.app

Flere episoder fra "Aarva"