Open Korean LLM - Ko-EQ-Bench: leaderboard
Korean EQ-Bench (Flitto): the model rates the intensity of emotions a character feels in a dialogue, scored against reference ratings. EQ-Bench points top out at 100 and go negative for badly calibrated answers; 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: EQ-Bench score. Source: huggingface.co. Saturation forecast: Estimated already saturated. 1158 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma-2-Ataraxy-Gemmasutra-9B-slerp | 62.53 | #584 |
| 2 | Gemma-2-9B-It-SPPO-Iter3 | 62.4 | #653 |
| 3 | recoilme-gemma-2-9B-v0.4 | 61.43 | #574 |
| 4 | Gemma-2-9B-It-SPPO-Iter2 | 61.4 | #627 |
| 5 | Gemma 2 9B (IT) | 60.47 | #795 |
| 6 | Gemmasutra-9B-v1 | 60.3 | #701 |
| 7 | Linkbricks-Horizon-AI-Korean-Superb-22B | 59.77 | #722 |
| 8 | qwen2.5-reinstruct-alternate-lumen-14B | 59.26 | #445 |
| 9 | Qwen2.5-14B-Gutenberg-1e-Delta | 59.25 | #473 |
| 10 | Qwen2.5-Lumen-14B | 59.23 | #482 |
| 11 | SuperNova-Medius | 59.13 | #483 |
| 12 | recoilme-gemma-2-9B-v0.2 | 58.92 | #651 |
| 13 | Qwen2.5-14B-Gutenberg-Instruct-Slerpeno | 58.84 | #431 |
| 14 | gemma-2-9B-it-SimPO | 58.77 | #569 |
| 15 | Mistral-Small-Instruct-2409-abliterated | 58.61 | #809 |
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-eq-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.