Configurable-Yi-1.5-9B-Chat: benchmark results
vicgalle's Configurable Safety Tuning fine-tune of Yi 1.5 9B Chat, letting system prompts set safety behavior (arXiv:2404.00495). Provider: vicgalle. Released 2024-05-12. Access: Open.
Aggregate Rating 1435 ± 16, rank #1074 of 1656 rated models, from 41 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 70.99 | Accuracy (%) (5-shot) | 94.5 |
| Open LLM Leaderboard v1 - GSM8K | 70.58 | Accuracy (%) (5-shot) | 93.6 |
| Open CoT - LogiQA | 7.99 | CoT Gain (%) | 88.9 |
| Open Chinese LLM - GSM8K | 62.32 | Accuracy (%) | 87.3 |
| Open Korean LLM - Ko-EQ-Bench | 36.28 | EQ-Bench score | 86.4 |
| Open CoT - LogiQA 2 | 13.74 | CoT Gain (%) | 85.5 |
| Open LLM Leaderboard - GPQA | 34.31 | Score | 85.4 |
| Open Chinese LLM Leaderboard | 63.99 | Official Nonzero-Task Mean (%) | 78.7 |
| Open CoT Leaderboard | 11.47 | Average CoT Gain (%) | 77.1 |
| Open Chinese LLM - ARC Challenge | 57.76 | Accuracy (%) | 76.6 |
| Open Korean LLM - KorNAT-Harmless | 64.7 | Normalized accuracy (%) | 76.4 |
| Open Chinese LLM - C-Eval Semantic | 84.47 | Accuracy (%) | 75.7 |
Interactive version: theaggregate.ai/model?slug=configurable-yi-1-5-9b-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.