Daredevil-8B-abliterated-dpomix: benchmark results
mlabonne's uncensored merge of Daredevil-8B, abliterated to remove refusals and then DPO fine-tuned on orpo-dpo-mix-40k to recover lost capability. Provider: mlabonne. Released 2024-06-01.
Aggregate Rating 1479 ± 16, rank #845 of 1635 rated models, from 26 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 71.8 | Accuracy (%) (5-shot) | 96.7 |
| Open Korean LLM Leaderboard | 47.1 | Average Score (%) | 94.5 |
| Open Korean LLM - Ko-IFEval | 49.63 | Strict accuracy, prompt/instruction mean (%) | 93.1 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 58.07 | Exact match, flexible extract (%) | 92.2 |
| Open LLM Leaderboard v1 - MMLU | 69.1 | Accuracy (%) (5-shot) | 92.2 |
| Open Medical LLM - PubMedQA | 77.4 | Accuracy (%) | 92.2 |
| Open Medical LLM - MMLU Medical Genetics | 84 | Accuracy (%) | 91.4 |
| Open Korean LLM - KorNAT-Helpful | 51.23 | Normalized accuracy (%) | 89.4 |
| Open Korean LLM - Ko-EQ-Bench | 37.5 | EQ-Bench score | 88.2 |
| Open Medical LLM - MedMCQA | 58.09 | Accuracy (%) | 87.5 |
| Open Medical-LLM Leaderboard (BenchmarkList) | 70.99 | Medical average (%) | 83.9 |
| Open Medical LLM | 70.99 | Average Accuracy (%) | 83.3 |
Interactive version: theaggregate.ai/model?slug=daredevil-8b-abliterated-dpomix · How It Works · Data refreshed daily, snapshot 2026-11-05.