Open Korean LLM - Ko-Winogrande: leaderboard
Korean WinoGrande: two-option commonsense referent-resolution sentences provided by Flitto, scored by accuracy on a 0-100 scale (50 is chance); 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: 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 | 76.09 | #1055 |
| 2 | nox-solar-10.7B-v4 | 73.88 | #1250 |
| 3 | free-solar-evo-v0.13 | 73.56 | #1077 |
| 4 | free-solar-evo-v0.1 | 73.01 | #1152 |
| 5 | free-solar-evo-v0.11 | 72.85 | #1144 |
| 6 | lambda-qwen2.5-14B-dpo-test | 71.03 | #481 |
| 7 | Rombos-LLM-V2.6-Qwen-14B | 70.24 | #520 |
| 8 | Qwen2.5-Lumen-14B | 70.24 | #482 |
| 9 | gemma-2-27B-it-SimPO-37K-100steps | 70.24 | #609 |
| 10 | mistral-nemo-wissenschaft-12B | 70.24 | #689 |
| 11 | gemma2-gutenberg-27B | 70.24 | #453 |
| 12 | Gemma 2 27B (IT) | 70.09 | #652 |
| 13 | SuperNova-Medius | 70.09 | #483 |
| 14 | Qwen2.5-14B-Gutenberg-1e-Delta | 69.93 | #473 |
| 15 | Mistral-Nemo-Instruct-2407-abliterated | 69.93 | #728 |
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-ko-winogrande · How It Works · Data refreshed daily, snapshot 2026-11-05.