Qwen2.5-7B-Instruct-abliterated: benchmark results
huihui-ai's original abliterated Qwen2.5 7B Instruct with refusals removed, first in a series later revised as v2 and v3. Provider: Alibaba. Released 2024-09-19.
Aggregate Rating 1498 ± 18, rank #790 of 1656 rated models, from 144 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 62.24 | Strict accuracy, prompt/instruction mean (%) | 97.1 |
| Open Korean LLM - KorNAT-Helpful | 56.28 | Normalized accuracy (%) | 96.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 63.61 | Exact match, flexible extract (%) | 96.2 |
| Open Korean LLM - Ko-EQ-Bench | 52.11 | EQ-Bench score | 96 |
| Open Korean LLM - KorNAT-Harmless | 72.27 | Normalized accuracy (%) | 95.4 |
| Open Portuguese LLM - ASSIN2 RTE | 94.12 | Macro F1 (%) | 95 |
| Open Korean LLM Leaderboard | 47.11 | Average Score (%) | 94.6 |
| Open Ko-LLM Leaderboard | 47.11 | Average (%) | 94.5 |
| Open LLM Leaderboard - MATH Level 5 | 45.77 | Score | 93.9 |
| Open Portuguese LLM - FaQuAD NLI | 81.54 | Macro F1 (%) | 92.3 |
| Open LLM Leaderboard - IFEval | 75.46 | Score | 90.9 |
| Open Portuguese LLM - ENEM | 74.32 | Accuracy (%) | 86.7 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-7b-instruct-abliterated · How It Works · Data refreshed daily, snapshot 2026-10-20.