Qwen2.5-7B-Instruct-Uncensored: benchmark results
Orion-zhen's uncensored SFT+DPO fine-tune of Qwen2.5 7B Instruct, trained to remove refusals while keeping base capabilities. Provider: Orion-zhen. Released 2024-09-26. Access: Open.
Aggregate Rating 1530 ± 17, rank #684 of 1656 rated models, from 144 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GSM8K (flexible extract) | 71.34 | Exact match, flexible extract (%) | 98.4 |
| Open Ko-LLM Leaderboard | 50.22 | Average (%) | 97.3 |
| Open Korean LLM Leaderboard | 50.22 | Average Score (%) | 97.3 |
| Open Korean LLM - Ko-IFEval | 56.66 | Strict accuracy, prompt/instruction mean (%) | 96.8 |
| Open LLM Leaderboard - MATH Level 5 | 47.73 | Score | 95.3 |
| Open Korean LLM - Ko-EQ-Bench | 48.93 | EQ-Bench score | 94.8 |
| Open Portuguese LLM - ASSIN2 RTE | 93.75 | Macro F1 (%) | 91.7 |
| Open Korean LLM - KorNAT-CKA | 32.82 | Normalized accuracy (%) | 89.5 |
| Open Portuguese LLM - ENEM | 75.44 | Accuracy (%) | 89.3 |
| Open Portuguese LLM - BLUEX | 66.62 | Accuracy (%) | 89 |
| Open Portuguese LLM - OAB Exams | 54.76 | Accuracy (%) | 87.1 |
| Open Arabic LLM - Arabic MMLU Arabic Language (Grammar) | 54.25 | Accuracy (%) | 85.8 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-7b-instruct-uncensored · How It Works · Data refreshed daily, snapshot 2026-10-20.