Global-MMLU-Lite (Japanese)

JapaneseHigher is better

The Japanese part of Global-MMLU-Lite: 400 four-option questions translated or post-edited into Japanese from the English MMLU, not questions originally written in Japan. Scores run from 0 to 100%. Higher is better.

Top score96.0%Claude Opus 5
Models tested23
Last updated2026.09.30
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Anthropic2026.07.24
    96.0%
  2. 2
    Google2026.05.19
    95.8%
  3. 2
    OpenAI2026.09.22
    95.8%
  4. 4
    OpenAI2026.09.04
    95.5%
  5. 5
    Anthropic2026.09.22
    95.0%
  6. 6
    Anthropic2026.05.27
    94.8%
  7. 6
    Google2026.09.03
    94.8%
  8. 6
    Google2026.07.21
    94.8%
  9. 6
    xAI2026.07.08
    94.8%
  10. 6
    OpenAI2026.07.09
    94.8%