Operational safety evaluation on OFFTOPICEVAL Chinese 1.0 (test)
99.32Acceptance Rate (ID)GPT-OSS (20B)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-OSS (20B)Family=GPT2025.09 | 99.32 | 55.37 | 35.57 | 62.38 | |
| Llama-3.3 (70B)Family=Llama-32025.09 | 99.23 | 55.27 | 7.98 | 47.96 | |
| Qwen-3 (235B)Family=Qwen2025.09 | 99.05 | 98.96 | 39.62 | 81.54 | |
| GPT-OSS (120B)Family=GPT2025.09 | 98.3 | 85.29 | 45.52 | 78.55 | |
| Llama-3.1 (8B)Family=Llama-32025.09 | 97.9 | 28.83 | 1.94 | 26.59 | |
| Phi-4 (15B)Family=Phi2025.09 | 92.76 | 79.83 | 34.52 | 50.11 | |
| Gemma-3 (12B)Family=Gemma2025.09 | 86.86 | 42.39 | 7.79 | 13.9 | |
| Gemma-3 (27B)Family=Gemma2025.09 | 79.24 | 50.47 | 12.58 | 22.26 | |
| Qwen-3 (30B)Family=Qwen2025.09 | 76.57 | 79.58 | 36.36 | 65.98 | |
| Mistral (24B)Family=Mistral2025.09 | 70.57 | 99.72 | 89.89 | 80.91 | |
| Phi-4-mini (4B)Family=Phi2025.09 | 67.14 | 37.68 | 85.84 | 58.34 | |
| Ministral (8B)Family=Mistral2025.09 | 64.29 | 60.31 | 4.88 | 43.26 |