Open Korean LLM - Ko-IFEval: leaderboard
Korean IFEval (Flitto): prompts with verifiable formatting and content instructions. The score is the mean of prompt-level and instruction-level strict accuracy, exactly as the Space computes its column, 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: Strict accuracy, prompt/instruction mean (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 1158 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemma2-gutenberg-27B | 74.68 | #453 |
| 2 | Gemma 2 27B (IT) | 72.48 | #652 |
| 3 | recoilme-gemma-2-9B-v0.2 | 69.49 | #651 |
| 4 | Qwen2.5-14B-Gutenberg-Instruct-Slerpeno | 68.42 | #431 |
| 5 | Gemma 2 9B (IT) | 68.42 | #795 |
| 6 | Rombos-LLM-V2.6-Qwen-14B | 68.27 | #520 |
| 7 | Qwen2.5-Lumen-14B | 68.22 | #482 |
| 8 | Qwen2.5-14B-Gutenberg-1e-Delta | 68.07 | #473 |
| 9 | SuperNova-Medius | 67.52 | #483 |
| 10 | Gemma-2-9B-It-SPPO-Iter3 | 67.39 | #653 |
| 11 | qwen2.5-reinstruct-alternate-lumen-14B | 67.07 | #445 |
| 12 | Qwen 2.5 14B Instruct | 66.99 | #654 |
| 13 | Gemma-2-9B-It-SPPO-Iter2 | 66.79 | #627 |
| 14 | lambda-qwen2.5-14B-dpo-test | 66.72 | #481 |
| 15 | Gemma2-9B-IT-Simpo-Infinity-Preference | 66.22 | #726 |
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-ifeval · How It Works · Data refreshed daily, snapshot 2026-11-05.