All Benchmarks/Safety & alignment

Horangi 4 — ALT (alignment)

KoreanHigher is better

Horangi 4's Alignment Performance (ALT) average: an equal-weight average of five areas in Korean, namely following instructions, ethical judgment, spotting hate speech, avoiding bias, and avoiding hallucination. It covers instruction-following and hallucination, not just safety. Scores run from 0 to 100%. Higher is better.

Other versions:HAE-RAE Bench v1 (Reading)Horangi 4 — GLP (general language)Horangi 4 (Korean LLM Leaderboard)HRM8KIFEval-KoKMMLUKMMLU-ProKo-AIME 2025Ko-ARC-AGIKo-HLEKo-MoralKo-MTBenchKo-TruthfulQAKoBALT-700 (semantics)KoBALT-700 (syntax)KoBBQ (bias)Korean Hate Speech
Top score84.8%Gemini 3.1 Pro
Models tested48
Last updated2026.08.03
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Google
    Gemini 3.1 Pro
    2026.02.19 · high-effort
    84.8%
  2. 2
    Alibaba
    Qwen3.7 Max
    2026.05.21
    81.6%
  3. 3
    Anthropic
    Claude Fable 5
    2026.06.09 · high-effort
    81.3%
  4. 4
    Google
    Gemini 3 Flash
    2025.12.17
    81.2%
  5. 5
    Z.ai
    GLM 5.2
    2026.06.16 · high-effort
    81.1%
  6. 6
    OpenAI
    GPT-5.5
    2026.04.23 · xhigh-effort
    80.8%
  7. 7
    Z.ai
    GLM-5-Turbo
    2026.03.15
    80.4%
  8. 8
    Google
    Gemini 3.5 Flash
    2026.05.19
    80.1%
  9. 9
    Anthropic
    Claude Opus 4.1
    2025.08.05
    80.0%
  10. 10
    Anthropic
    Claude Opus 4
    2025.05.22 · high-effort
    79.6%

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.