Open Korean LLM - Ko-GPQA-Diamond: leaderboard
Ko-GPQA Diamond: a Korean version of GPQA Diamond's graduate-level biology, physics and chemistry multiple-choice questions provided by Flitto, run zero-shot and scored by normalized accuracy on a 0-100 scale (25 is four-option 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: 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 | 29.8 | #1055 |
| 2 | free-solar-evo-v0.13 | 29.29 | #1077 |
| 3 | L3.1-Niitorm-8B-DPO-t0.0001 | 29.29 | #826 |
| 4 | Qwen2.5-14B-Gutenberg-1e-Delta | 28.79 | #473 |
| 5 | gemma-2-9B-it-SimPO | 28.79 | #569 |
| 6 | bagel-8B-v1.0 | 28.79 | #1031 |
| 7 | Infinity-Instruct-7M-Gen-Llama3.1-8B | 28.79 | #857 |
| 8 | Mistral7B-PairRM-SPPO-ExPO | 28.79 | #1332 |
| 9 | Mistral-7B-Base-SFT-RDPO | 28.79 | #1333 |
| 10 | MagpieLM-8B-Chat-v0.1 | 28.28 | #1192 |
| 11 | Gemma 2 27B (IT) | 27.78 | #652 |
| 12 | Qwen2.5-Lumen-14B | 27.78 | #482 |
| 13 | free-solar-evo-v0.11 | 27.78 | #1144 |
| 14 | Mistral7B-PairRM-SPPO | 27.78 | #1284 |
| 15 | Qwen2.5-14B-Gutenberg-Instruct-Slerpeno | 27.78 | #431 |
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-gpqa-diamond · How It Works · Data refreshed daily, snapshot 2026-11-05.