Benchmark
对齐可靠榜
security
text
多语言
对齐可靠榜：多维度评测综合成绩。评测分数为事实数据，整理自公开评测体系，仅供学习研究参考。
Top 10
榜单概览
数据说明
分数为 0-100 百分制综合成绩（Elo 榜为相对评分，仅供同榜内比较）。评测分数属事实数据，整理自公开评测体系与厂商公开资料，部分分数为厂商自报口径，不同榜单的评测方法与数据集可能存在差异，请结合多个榜单交叉参考。
Ranking
模型排名
| 名次 | 模型 | 机构 | 主分 | Fairness | Individual Harm | Legality | Privacy | Civic Virtue |
|---|---|---|---|---|---|---|---|---|
| 1 | GPT-3.5 Turbo | OpenAI | 96.0 | 86.7 | 100.0 | 100.0 | 100.0 | 93.3 |
| 2 | Claude | Anthropic | 85.3 | 86.7 | 73.3 | 86.7 | 93.3 | 86.7 |
| 3 | Qwen-14B | Alibaba Cloud / Qwen Team | 69.3 | 73.3 | 73.3 | 53.3 | 73.3 | 73.3 |
| 4 | InternLM-20B | Shanghai AI Lab | 69.3 | 66.7 | 80.0 | 53.3 | 66.7 | 80.0 |
| 5 | Vicuna-13B-v1.5 | LMSYS | 58.7 | 60.0 | 60.0 | 33.3 | 60.0 | 80.0 |
| 6 | Vicuna-33B-v1.3 | LMSYS | 57.3 | 66.7 | 40.0 | 60.0 | 60.0 | 60.0 |
| 7 | InternLM-7B | Shanghai AI Lab | 57.3 | 53.3 | 66.7 | 46.7 | 46.7 | 73.3 |
| 8 | Qwen-7B | Alibaba Cloud / Qwen Team | 54.7 | 46.7 | 73.3 | 33.3 | 46.7 | 73.3 |
| 9 | Baichuan2-13B | Baichuan AI | 45.3 | 53.3 | 40.0 | 26.7 | 33.3 | 73.3 |
| 10 | ChatGLM3-6B | Zhipu AI | 45.3 | 46.7 | 53.3 | 20.0 | 40.0 | 66.7 |
| 11 | Vicuna-7B-v1.5 | LMSYS | 25.3 | 33.3 | 20.0 | 6.7 | 26.7 | 40.0 |
| 12 | Baichuan2-7B | Baichuan AI | 20.0 | 26.7 | 13.3 | 6.7 | 20.0 | 33.3 |
| 13 | ChatGLM2-6B | Zhipu AI | 17.3 | 20.0 | 20.0 | 0.0 | 6.7 | 40.0 |
| 14 | MOSS-SFT | Fudan University | 10.7 | 13.3 | 13.3 | 13.3 | 13.3 | 0.0 |