All Benchmarks/Instruction following

Ko-MTBench

KoreanHigher is better

80 questions, each answered over two turns, across eight areas: writing, roleplay, reasoning, math, coding, extraction, STEM, and humanities. It adapts MT-Bench to Korean language and culture; an AI judge scores answers from 1 to 10, divided by 10. Scores run from 0 to 100%. Higher is better.

Other versions:HAE-RAE Bench v1 (Reading)Horangi 4 — ALT (alignment)Horangi 4 — GLP (general language)Horangi 4 (Korean LLM Leaderboard)HRM8KIFEval-KoKMMLUKMMLU-ProKo-AIME 2025Ko-ARC-AGIKo-HLEKo-MoralKo-TruthfulQAKoBALT-700 (semantics)KoBALT-700 (syntax)KoBBQ (bias)Korean Hate Speech
Top score83.5%Motif 3
Models tested48
Last updated2026.08.03
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Motif Technologies
    Motif 3
    2026.08.10
    83.5%
  2. 2
    Google
    Gemini 3 Flash
    2025.12.17
    9.0%
  3. 2
    Google
    Gemini 3.5 Flash
    2026.05.19
    9.0%
  4. 2
    Anthropic
    Claude Opus 4.5
    2025.11.24 · high-effort
    9.0%
  5. 5
    Anthropic
    Claude Sonnet 4.6
    2026.02.17 · high-effort
    8.9%
  6. 6
    Anthropic
    Claude Sonnet 4.5
    2025.09.29 · high-effort
    8.9%
  7. 7
    Anthropic
    Claude Sonnet 5
    2026.06.30 · high-effort
    8.9%
  8. 7
    OpenAI
    GPT OSS 120B
    2025.08.05 · high-effort
    8.9%
  9. 7
    Google
    Gemini 3.1 Pro
    2026.02.19 · high-effort
    8.9%
  10. 10
    Alibaba
    Qwen3.7 Max
    2026.05.21
    8.9%

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.