Open Korean LLM - KorNAT-CKA: leaderboard
KorNAT common knowledge alignment (CKA, by SELECTSTAR and KAIST AI): multiple-choice questions on the general knowledge taught in Korean schools, scored by normalized accuracy on a 0-100 scale; higher is better. From Upstage and NIA's Open Ko-LLM Leaderboard 2 (lm-evaluation-harness on private Korean test sets); frozen since its last evaluations in January 2025.
Metric: Normalized accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 1158 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Myrrh_solar_10.7b_3.0 | 48.57 | #1055 |
| 2 | lambda-qwen2.5-14B-dpo-test | 46.25 | #481 |
| 3 | Qwen2.5-Lumen-14B | 40.1 | #482 |
| 4 | Qwen2.5-14B-Gutenberg-Instruct-Slerpeno | 39.85 | #431 |
| 5 | free-solar-evo-v0.13 | 39.81 | #1077 |
| 6 | Qwen2.5-14B-Gutenberg-1e-Delta | 39.68 | #473 |
| 7 | qwen2.5-reinstruct-alternate-lumen-14B | 39.66 | #445 |
| 8 | Qwen 2.5 14B Instruct | 39.56 | #654 |
| 9 | recoilme-gemma-2-9B-v0.4 | 39.31 | #574 |
| 10 | Rombos-LLM-V2.6-Qwen-14B | 39.21 | #520 |
| 11 | NeuralSynthesis-7B-v0.1 | 38.9 | #973 |
| 12 | free-solar-evo-v0.11 | 38.47 | #1144 |
| 13 | gemma2-gutenberg-27B | 38.05 | #453 |
| 14 | SuperNova-Medius | 37.8 | #483 |
| 15 | Gemma 2 27B (IT) | 37.57 | #652 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-korean-llm-kornat-cka · How It Works · Data refreshed daily, snapshot 2026-11-05.