Best AI for your language

The same text costs more in some languages than in others, and the difference depends on the model. Pick your language to see the measured cost and the best models for it.

Choose a language

Tokens compared with English, by language and model family

LanguageGPTGemmaLlamaQwenDeepSeekMistralGLMAverage
Chinese 中文+11%−1%−7%−11%−16%+40%−12%+0%
Spanish Español+17%+17%+19%+38%+45%+25%+28%+27%
Portuguese Português+15%+20%+17%+44%+47%+30%+37%+30%
German Deutsch+17%+21%+19%+45%+50%+30%+30%+30%
Russian Русский+19%+14%+6%+60%+51%+45%+21%+31%
French Français+25%+32%+27%+48%+49%+27%+39%+35%
Indonesian Bahasa Indonesia+31%+18%+19%+66%+47%+50%+65%+42%
Japanese 日本語+61%+17%+33%+41%+54%+70%+53%+47%
Italian Italiano+46%+32%+28%+69%+58%+47%+54%+48%
Korean 한국어+39%+30%+20%+62%+66%+30%+89%+48%
Vietnamese Tiếng Việt+45%+36%+34%+42%+124%+52%+45%+54%
Turkish Türkçe+35%+30%+32%+56%+103%+57%+67%+54%
Arabic العربية+35%+47%+58%+64%+68%+26%+87%+55%
Hindi हिन्दी+57%+30%+69%+380%+201%+91%+413%+177%

Tokens needed for the same four texts, relative to English, measured with each company’s public tokenizer. A minus sign means cheaper than English. Last update: Sep 25, 2026 · How we work