Technology · 2026-10-27 · Afternoon briefing

Study: AI costs up to 17 times more for non-English users

Using AI models in languages other than English can be significantly more expensive, a phenomenon dubbed the 'tokenizer tax'. A study by the Coalition for Language Equity in AI Research found some languages require 1.5 to 5 times more processing 'tokens' than the cheapest. For one model tested across 200 languages, the cost gap grew to 17 times.

The disparity has direct commercial consequences: the study notes a Bengaluru startup could face a surcharge its Austin competitor does not. Beyond cost, higher token counts also mean slower responses and reduced effective context capacity. Pratyush Choudhury of Activate AI, a venture fund dedicated to AI, likens it to a tutor forgetting a student's history faster in one language, degrading the lesson's quality.

While token prices are falling, Dheemanth Reddy of Maya Research argues labs must deliberately redesign their systems to handle other languages better. Pratyush Choudhury calls for major labs to publish 'language-by-language scorecards' and commit to measurable improvements for their worst-served users.

Summary written by Cup of News with AI from the sources below. Check the originals for the full story.

Read the sources

• AI models cost more when you don’t use English
• Costs tied to No Surprises Act's IDR process hit $22B in 2025, study finds

More stories in this edition

1. EU unveils climate law to prepare for a 3°C warmer world
2. Malawi's Chawinga is first African woman in Ballon d'Or top 10
3. India’s largest IPO may end its cheap data era
4. Ukraine and Russia begin local ceasefire in occupied Oleshky
5. Kenya releases Ebola contacts as DRC trials vaccines
6. Harry Kane wins Ballon d'Or, ending German football's long wait
7. OpenAI, Anthropic claim breakthroughs in dueling proofs
8. Nordic leaders in rare call with Putin over Ukraine talks
9. Kaepernick has no regrets over career-ending protest, ten years on

Read the whole edition