Privacy Sanitization on PRIVASIS-SANITIZATION Hard (test)
80.2Sanitization Success (Attribute)o3
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| o32026.02 | 80.2 | 75.65 | 15.23 | 87.89 | 87.04 | 83.81 | 11.66 | |
| GPT-52026.02 | 78.78 | 75.3 | 16.28 | 87.08 | 87.23 | 84.25 | 13.14 | |
| GPT-OSSParameters=120B2026.02 | 77.67 | 74.07 | 13.84 | 88.36 | 87.87 | 84.94 | 10.53 | |
| LLaMA-4 Maverick2026.02 | 76.21 | 73.4 | 16.19 | 82.07 | 81.92 | 78.24 | 11.05 | |
| DeepSeek R12026.02 | 75.54 | 72.76 | 15.14 | 86.7 | 86.16 | 82.59 | 11.23 | |
| GPT-4.12026.02 | 75.14 | 72.4 | 13.93 | 89.79 | 90.06 | 86.51 | 12.18 | |
| PRIVASIS-CLEANERParameters=4B2026.02 | 73.29 | 71.23 | 14.19 | 95.76 | 95.6 | 92.96 | 12.36 | |
| Qwen3Parameters=235B2026.02 | 69.01 | 67.33 | 12.79 | 89.37 | 89.71 | 86.95 | 10.27 | |
| PRIVASIS-CLEANERParameters=0.6B2026.02 | 68.11 | 66.14 | 11.31 | 95.09 | 95.2 | 91.99 | 9.31 | |
| Qwen3Parameters=4B2026.02 | 62.15 | 59.89 | 8.27 | 84.36 | 84.01 | 80.68 | 5.66 | |
| Qwen3Parameters=0.6B2026.02 | 33.41 | 35.46 | 12.53 | 76.63 | 76.08 | 72.58 | 4.44 |