Meta-Llama-3.1-8B-Instruct-abliterated: benchmark results
mlabonne's abliterated Llama 3.1 8B Instruct, an uncensored version made with the abliteration technique developed by FailSpy. Provider: mlabonne. Released 2024-07-24. Access: Open.
Aggregate Rating 1412 ± 21, rank #1179 of 1654 rated models, from 20 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 45.59 | Strict accuracy, prompt/instruction mean (%) | 90 |
| Open LLM Leaderboard - IFEval | 73.29 | Score | 87.7 |
| Open Korean LLM - Ko-EQ-Bench | 33.12 | EQ-Bench score | 82.9 |
| Open Korean LLM - KorNAT-Helpful | 49.7 | Normalized accuracy (%) | 81 |
| Open Korean LLM - Ko-GPQA-Diamond | 25.25 | Normalized accuracy (%) | 75.9 |
| Open Portuguese LLM - ASSIN2 RTE | 92.15 | Macro F1 (%) | 74.9 |
| Open Portuguese LLM - FaQuAD NLI | 75.45 | Macro F1 (%) | 73.9 |
| Open Portuguese LLM - OAB Exams | 48.15 | Accuracy (%) | 68.4 |
| Open Korean LLM Leaderboard | 38.58 | Average Score (%) | 67.6 |
| Open Portuguese LLM - ENEM | 66.06 | Accuracy (%) | 63.5 |
| Open Portuguese LLM - BLUEX | 53.41 | Accuracy (%) | 55.9 |
| Open LLM Leaderboard - MMLU-Pro | 35.03 | Score | 51.5 |
Interactive version: theaggregate.ai/model?slug=meta-llama-3-1-8b-instruct-abliterated · How It Works · Data refreshed daily, snapshot 2026-10-24.