All Benchmarks/Reasoning

Ko-ARC-AGI

KoreanHigher is better

ARC-AGI grid puzzles (infer a transformation rule from examples and draw the output for a new input) given with Korean instructions. The output grid must match exactly in size and values. Scores are not comparable with official ARC Prize results. Horangi runs a fixed 100-question subset rather than the full dataset, so its numbers are not directly comparable with full-dataset scores. 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-HLEKo-MoralKo-MTBenchKo-TruthfulQAKoBALT-700 (semantics)KoBALT-700 (syntax)KoBBQ (bias)Korean Hate Speech
Top score97.0%GPT-5.6 Terra
Models tested48
Last updated2026.08.03
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    OpenAI
    GPT-5.6 Terra
    2026.07.09 · max-effort
    97.0%
  2. 1
    OpenAI
    GPT-5.6 Sol
    2026.07.09 · max-effort
    97.0%
  3. 3
    Anthropic
    Claude Fable 5
    2026.06.09 · high-effort
    96.0%
  4. 4
    OpenAI
    GPT-5.4
    2026.03.05 · xhigh-effort
    95.0%
  5. 5
    OpenAI
    GPT-5.5
    2026.04.23 · xhigh-effort
    93.0%
  6. 6
    Google
    Gemini 3.1 Pro
    2026.02.19 · high-effort
    92.0%
  7. 7
    Motif Technologies
    Motif 3
    2026.08.10
    90.9%
  8. 8
    Google
    Gemini 3.5 Flash
    2026.05.19
    89.9%
  9. 9
    Anthropic
    Claude Sonnet 5
    2026.06.30 · high-effort
    85.0%
  10. 10
    xAI
    Grok 4.5
    2026.07.08
    84.0%

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.