SuperNova-Medius: benchmark results
Arcee AI's 14B Qwen2.5-based model distilled from Llama 3.1 405B and Qwen2.5 72B teachers. Provider: Arcee AI. Released 2024-10-02. Access: Open.
Aggregate Rating 1589 ± 18, rank #479 of 1656 rated models, from 27 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Ko-LLM Leaderboard | 54.26 | Average (%) | 99.5 |
| Open Korean LLM Leaderboard | 54.26 | Average Score (%) | 99.5 |
| Open Korean LLM - Ko-IFEval | 67.52 | Strict accuracy, prompt/instruction mean (%) | 99 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 75.21 | Exact match, flexible extract (%) | 98.9 |
| Open Korean LLM - Ko-EQ-Bench | 59.13 | EQ-Bench score | 98.4 |
| Open Portuguese LLM - ASSIN2 RTE | 94.65 | Macro F1 (%) | 97.6 |
| Open Korean LLM - KorNAT-CKA | 37.8 | Normalized accuracy (%) | 97.4 |
| Open Korean LLM - KorNAT-Helpful | 58.63 | Normalized accuracy (%) | 96.9 |
| Open Korean LLM - Ko-Winogrande | 70.09 | Accuracy (%) | 96.4 |
| Open LLM Leaderboard - MATH Level 5 | 46.9 | Score | 94.8 |
| Open Korean LLM - KorNAT-Harmless | 70.63 | Normalized accuracy (%) | 93.7 |
| Open Portuguese LLM - BLUEX | 72.04 | Accuracy (%) | 93.1 |
Interactive version: theaggregate.ai/model?slug=supernova-medius · How It Works · Data refreshed daily, snapshot 2026-10-20.