Horangi 4 - KMMLU-Pro: leaderboard
KMMLU-Pro, multiple-choice questions from Korean national professional licensure exams, as sampled by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Accuracy (%), higher is better.
Metric: Accuracy (%). Source: horangi.ai. Saturation forecast: Estimated already saturated. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 95 | #22 (GPT-5.6 Sol) |
| 2 | Claude Opus 4.8 (High) | 95 | #39 (Claude Opus 4.8) |
| 3 | GPT-5.5 (xHigh) | 94 | #33 (GPT-5.5) |
| 4 | GPT-5.6 Terra (Max) | 91 | #40 (GPT-5.6 Terra) |
| 5 | Claude Opus 4.6 (High) | 91 | #67 (Claude Opus 4.6) |
| 6 | GPT-5.4 (2026-03-05) (xHigh) | 91 | #63 (GPT-5.4 (2026-03-05)) |
| 7 | GPT-5.6 Luna (Max) | 90 | #79 (GPT-5.6 Luna) |
| 8 | Claude Sonnet 5 (High) | 90 | #65 (Claude Sonnet 5) |
| 9 | Gemini 3 Pro (Preview) (High) | 90 | #71 (Gemini 3 Pro (Preview)) |
| 10 | Gemini 3.5 Flash | 89.9 | #66 |
| 11 | Gemini 3.1 Pro (Preview) (High) | 89.9 | #64 (Gemini 3.1 Pro (Preview)) |
| 12 | Claude Sonnet 4.6 (High) | 89 | #94 (Claude Sonnet 4.6) |
| 13 | GPT-5.2 (2025-12-11) (xHigh) | 88.66 | |
| 14 | Grok 4.5 | 88 | #54 |
| 15 | Claude Fable 5 (High) | 88 | #20 (Claude Fable 5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=horangi-4-kmmlu-pro · How It Works · Data refreshed daily, snapshot 2026-11-05.