Linkbricks-Horizon-AI-Korean-Superb-22B: benchmark results
Linkbricks Horizon-AI's Korean Superb 22B, built on Mistral Small (2409) with about 45% of parameters retrained via Korean continued-pretraining, SFT and DPO on a 90M-article news corpus. Provider: Linkbricks Horizon-AI. Released 2024-11-08. Access: Open.
Aggregate Rating 1518 ± 18, rank #722 of 1635 rated models, from 65 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-EQ-Bench | 59.77 | EQ-Bench score | 99 |
| Open Korean LLM Leaderboard | 49.37 | Average Score (%) | 96.7 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 61.03 | Exact match, flexible extract (%) | 94.7 |
| Open Korean LLM - Ko-IFEval | 51.58 | Strict accuracy, prompt/instruction mean (%) | 94.6 |
| Open Japanese LLM - Wiki Dependency SET F1 | 43.14 | Score (%) | 90.9 |
| Open Japanese LLM - Xlsum JA RougeLsum | 26.79 | Score (%) | 89.3 |
| Open Japanese LLM - Xlsum JA Rouge1 | 31.71 | Score (%) | 87.7 |
| Open Korean LLM - Ko-Winogrande | 65.59 | Accuracy (%) | 85.3 |
| Open Japanese LLM - Xlsum JA Bert Score JA F1 | 70.33 | Score (%) | 84.1 |
| Open Japanese LLM - Mbpp Pylint Check | 92.57 | Score (%) | 82.1 |
| Open Japanese LLM - SUM | 11.04 | Score (%) | 81.7 |
| Open Japanese LLM - Xlsum JA Rouge2 | 11.04 | Score (%) | 81.5 |
Interactive version: theaggregate.ai/model?slug=linkbricks-horizon-ai-korean-superb-22b · How It Works · Data refreshed daily, snapshot 2026-11-05.