Qwen2.5-Lumen-14B: benchmark results
v000000's preference-tuned (DPO) fine-tune of Qwen2.5-14B-Instruct targeting prompt adherence, storywriting, and roleplay. Provider: v000000. Released 2024-09-20. Access: Open.
Aggregate Rating 1589 ± 17, rank #477 of 1656 rated models, from 139 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Ko-LLM Leaderboard | 54.31 | Average (%) | 99.6 |
| Open Korean LLM Leaderboard | 54.31 | Average Score (%) | 99.6 |
| Open Korean LLM - Ko-IFEval | 68.22 | Strict accuracy, prompt/instruction mean (%) | 99.3 |
| Open Korean LLM - KorNAT-CKA | 40.1 | Normalized accuracy (%) | 99.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 75.82 | Exact match, flexible extract (%) | 99.1 |
| Open LLM Leaderboard - MATH Level 5 | 53.63 | Score | 98.6 |
| Open Korean LLM - Ko-EQ-Bench | 59.23 | EQ-Bench score | 98.5 |
| Open Korean LLM - KorNAT-Helpful | 60.9 | Normalized accuracy (%) | 98.2 |
| Open LLM Leaderboard - IFEval | 80.64 | Score | 98.1 |
| Open Korean LLM - Ko-Winogrande | 70.24 | Accuracy (%) | 96.8 |
| Open Korean LLM - KorNAT-Harmless | 73.37 | Normalized accuracy (%) | 96.5 |
| Open Korean LLM - Ko-GPQA-Diamond | 27.78 | Normalized accuracy (%) | 95 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-lumen-14b · How It Works · Data refreshed daily, snapshot 2026-10-20.