Benchmark
安全防御榜
security
text
多语言
安全防御榜：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 | Toxicity | Malicious Use | Child Safety | Info Safety | Socioeconomic | Bias | Human Rights |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Phi 4 | Microsoft | 96.8 | 98.0 | 97.5 | 96.4 | 96.2 | 96.9 | 96.6 | 96.1 |
| 2 | Claude-4-Sonnet | Anthropic | 95.3 | 95.4 | 95.0 | 89.8 | 94.8 | 96.4 | 99.4 | 96.5 |
| 3 | Phi-4-Mini-Instruct | Microsoft | 94.8 | 97.5 | 96.0 | 92.3 | 96.1 | 95.7 | 91.0 | 94.8 |
| 4 | GPT-4 Turbo | OpenAI | 93.8 | 96.9 | 94.4 | 91.5 | 94.2 | 91.4 | 93.6 | 94.8 |
| 5 | o1-mini | OpenAI | 93.5 | 93.9 | 94.9 | 90.0 | 97.1 | 91.2 | 94.0 | 93.7 |
| 6 | Claude 3.5 Sonnet | Anthropic | 93.3 | 96.3 | 91.9 | 88.0 | 93.0 | 91.1 | 97.3 | 95.6 |
| 7 | Falcon3-7B-Instruct | TII | 93.1 | 95.6 | 93.8 | 90.4 | 93.8 | 92.3 | 92.7 | 93.2 |
| 8 | Claude 3.7 Sonnet | Anthropic | 92.7 | 96.1 | 91.8 | 85.8 | 93.0 | 90.5 | 97.0 | 94.9 |
| 9 | o3 | OpenAI | 92.4 | 91.2 | 93.2 | 85.7 | 94.8 | 92.3 | 97.0 | 92.2 |
| 10 | o4-mini | OpenAI | 91.9 | 90.8 | 93.3 | 84.9 | 95.1 | 92.6 | 94.9 | 91.9 |
| 11 | GPT-5 | OpenAI | 91.3 | 92.3 | 92.0 | 87.3 | 94.2 | 93.8 | 92.6 | 87.2 |
| 12 | Falcon3-10B-Instruct | TII | 91.2 | 94.8 | 91.8 | 86.6 | 93.5 | 88.0 | 92.1 | 91.7 |
| 13 | Phi-3.5-mini-instruct | Microsoft | 90.2 | 95.8 | 88.9 | 84.8 | 92.0 | 88.6 | 90.3 | 90.6 |
| 14 | GPT-4.1 | OpenAI | 90.1 | 92.5 | 91.6 | 84.3 | 93.7 | 86.0 | 93.1 | 89.9 |
| 15 | Qwen3-14B | Alibaba Cloud / Qwen Team | 89.7 | 94.3 | 90.4 | 84.1 | 91.8 | 87.1 | 89.2 | 90.7 |
| 16 | GPT-4.1 nano | OpenAI | 89.6 | 94.2 | 90.6 | 85.1 | 92.7 | 87.6 | 87.8 | 89.3 |
| 17 | GPT-4o | OpenAI | 89.4 | 93.4 | 89.6 | 85.3 | 91.5 | 82.2 | 92.4 | 91.5 |
| 18 | Qwen3 32B | Alibaba Cloud / Qwen Team | 89.3 | 93.0 | 89.5 | 85.2 | 92.2 | 86.4 | 89.0 | 89.8 |
| 19 | Qwen2.5 14B Instruct | Alibaba Cloud / Qwen Team | 89.1 | 93.1 | 89.1 | 83.6 | 92.2 | 87.3 | 89.9 | 88.8 |
| 20 | Qwen Max | Alibaba Cloud / Qwen Team | 88.8 | 94.4 | 88.0 | 80.9 | 90.1 | 86.7 | 92.3 | 89.2 |
| 21 | DeepSeek-R1 | DeepSeek | 88.3 | 87.8 | 88.9 | 86.1 | 94.5 | 82.1 | 91.3 | 87.3 |
| 22 | Qwen2.5 32B Instruct | Alibaba Cloud / Qwen Team | 88.3 | 93.9 | 88.6 | 82.4 | 89.9 | 84.4 | 91.4 | 87.5 |
| 23 | GPT-4.1 mini | OpenAI | 88.1 | 91.7 | 87.3 | 81.9 | 90.0 | 84.9 | 95.3 | 85.4 |
| 24 | Qwen3-8B (Thinking) | Alibaba Cloud / Qwen Team | 87.9 | 93.1 | 88.1 | 79.8 | 91.9 | 85.5 | 87.7 | 89.3 |
| 25 | Qwen2.5 72B Instruct | Alibaba Cloud / Qwen Team | 87.1 | 93.1 | 86.2 | 81.5 | 88.0 | 83.3 | 89.5 | 88.2 |
| 26 | InternLM3-8B-Instruct | Shanghai AI Lab | 86.5 | 93.7 | 85.3 | 78.1 | 87.8 | 83.5 | 88.9 | 88.0 |
| 27 | DeepSeek-V3 | DeepSeek | 85.8 | 89.3 | 88.8 | 79.4 | 89.8 | 78.5 | 90.6 | 84.2 |
| 28 | Gemini-2.5-Flash-Preview | 85.6 | 90.1 | 82.7 | 71.5 | 89.9 | 82.4 | 95.9 | 87.0 | |
| 29 | Gemini-2.5-Pro-Preview | 83.2 | 84.6 | 79.2 | 77.2 | 89.7 | 72.9 | 94.3 | 84.5 | |
| 30 | Llama 3.1 8B Instruct | Meta | 83.0 | 87.8 | 84.4 | 77.4 | 88.2 | 83.1 | 81.2 | 79.2 |
| 31 | Llama 3.3 70B Instruct | Meta | 82.9 | 88.7 | 81.0 | 77.2 | 88.0 | 78.0 | 88.0 | 79.4 |
| 32 | Gemma 3 12B | 82.8 | 92.9 | 81.2 | 73.4 | 83.8 | 79.0 | 87.9 | 81.6 | |
| 33 | Yi-1.5-34B-Chat | 01.AI | 82.8 | 92.3 | 81.2 | 73.5 | 86.7 | 75.3 | 87.6 | 83.3 |
| 34 | Qwen2.5 7B Instruct | Alibaba Cloud / Qwen Team | 82.2 | 90.2 | 80.4 | 75.4 | 84.3 | 75.0 | 86.8 | 83.1 |
| 35 | Gemma 3 27B | 81.3 | 91.9 | 78.6 | 73.1 | 82.3 | 78.2 | 85.7 | 79.0 | |
| 36 | Grok-3 Mini | xAI | 80.7 | 85.7 | 79.3 | 66.8 | 93.9 | 68.7 | 91.3 | 79.7 |
| 37 | Hunyuan-A13B-Instruct | Tencent | 78.3 | 87.1 | 74.8 | 67.2 | 83.0 | 71.9 | 83.5 | 80.3 |
| 38 | Mistral-Large-Instruct-2411 | Mistral AI | 76.7 | 86.6 | 73.4 | 65.8 | 83.9 | 65.3 | 86.0 | 75.7 |
| 39 | MiniCPM4-8B | OpenBMB | 75.3 | 72.7 | 74.9 | 71.7 | 80.1 | 80.8 | 72.6 | 74.3 |
| 40 | Grok-4 | xAI | 74.8 | 80.9 | 73.0 | 60.8 | 88.0 | 60.3 | 87.7 | 72.5 |
| 41 | Mistral-Small-24B-Instruct | Mistral AI | 74.5 | 84.6 | 70.7 | 61.1 | 80.9 | 64.7 | 83.4 | 76.0 |
| 42 | Yi-1.5-9B-Chat | 01.AI | 73.2 | 83.2 | 71.9 | 64.9 | 77.0 | 64.6 | 78.6 | 71.7 |
| 43 | InternLM2.5-20B-Chat | Shanghai AI Lab | 72.0 | 63.0 | 62.7 | 74.7 | 70.6 | 86.7 | 74.7 | 72.0 |
| 44 | InternLM2.5-7B-Chat | Shanghai AI Lab | 72.0 | 61.9 | 65.0 | 72.7 | 70.9 | 88.9 | 73.3 | 71.0 |
| 45 | Llama 3.1 70B Instruct | Meta | 71.9 | 78.3 | 67.9 | 60.9 | 78.9 | 63.9 | 80.9 | 72.3 |
| 46 | GLM-4-32B-0414 | Zhipu AI | 69.7 | 61.6 | 61.8 | 67.5 | 69.6 | 87.0 | 70.7 | 69.4 |
| 47 | GLM-4-9B-0414 | Zhipu AI | 69.5 | 60.5 | 60.7 | 66.7 | 69.7 | 84.4 | 73.7 | 71.1 |
| 48 | Moonlight-16B-A3B-Instruct | Moonshot AI | 66.6 | 77.0 | 60.3 | 54.0 | 76.7 | 50.9 | 81.8 | 65.1 |
| 49 | GLM-Z1-9B-0414 | Zhipu AI | 62.1 | 54.3 | 53.8 | 63.0 | 61.3 | 74.1 | 64.8 | 59.9 |
| 50 | GLM-Z1-32B-0414 | Zhipu AI | 61.9 | 54.7 | 54.6 | 62.7 | 60.0 | 77.0 | 63.8 | 60.8 |