All Benchmarks/Multilingual

Global-MMLU-Lite (Korean)

KoreanHigher is better

The Korean part of Global-MMLU-Lite: 400 four-option questions translated or post-edited into Korean from the English MMLU, unlike KMMLU, which is built from Korean exams. Scores run from 0 to 100%. Higher is better.

Other versions:Global-MMLU-LiteGlobal-MMLU-Lite (Japanese)
Top score95.5%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
    Anthropic
    Claude Opus 5
    2026.07.24
    95.5%
  2. 1
    OpenAI
    GPT-6 Astra
    2026.09.04
    95.5%
  3. 3
    Anthropic
    Claude Opus 5.5
    2026.09.22
    95.3%
  4. 3
    Google
    Gemini 3.5 Flash
    2026.05.19
    95.3%
  5. 3
    Google
    Gemini 3.8 Flash
    2026.09.03
    95.3%
  6. 6
    Anthropic
    Claude Opus 4.1
    2025.08.05
    95.0%
  7. 7
    Google
    Gemini 3.7 Flash
    2026.08.13
    94.8%
  8. 8
    Anthropic
    Claude Opus 4.8
    2026.05.27
    94.5%
  9. 8
    OpenAI
    GPT-6 Sol
    2026.09.22
    94.5%
  10. 8
    Google
    Gemini 3.6 Flash
    2026.07.21
    94.5%