Daredevil-8B-abliterated: benchmark results
mlabonne's abliterated version of his Daredevil-8B Llama 3 merge, with the refusal direction removed for uncensored use. Provider: mlabonne. Released 2024-06-01. Access: Open.
Aggregate Rating 1481 ± 14, rank #846 of 1656 rated models, from 53 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 73.16 | Accuracy (%) (5-shot) | 98.6 |
| Open CoT - LSAT Reading Comprehension | 23.42 | CoT Gain (%) | 96.2 |
| Open Korean LLM - Ko-IFEval | 52.25 | Strict accuracy, prompt/instruction mean (%) | 95.1 |
| Open Korean LLM Leaderboard | 47.57 | Average Score (%) | 95 |
| Open Ko-LLM Leaderboard | 47.57 | Average (%) | 94.9 |
| Open CoT Leaderboard | 14.25 | Average CoT Gain (%) | 93.9 |
| Open CoT - LogiQA | 9.42 | CoT Gain (%) | 93.5 |
| Open Korean LLM - Ko-EQ-Bench | 43.09 | EQ-Bench score | 92.9 |
| Open CoT - LogiQA 2 | 15.01 | CoT Gain (%) | 92.7 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 58.61 | Exact match, flexible extract (%) | 92.7 |
| Open LLM Leaderboard v1 - MMLU | 68.99 | Accuracy (%) (5-shot) | 92.1 |
| Open Medical LLM - MMLU Medical Genetics | 84 | Accuracy (%) | 91.4 |
Interactive version: theaggregate.ai/model?slug=daredevil-8b-abliterated · How It Works · Data refreshed daily, snapshot 2026-10-20.