Safety content moderation on GUARDZOO in-domain (test)
89.21Safety ACC (Safe/Unsafe)ROUTEGUARD-4B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ROUTEGUARD-4BEvaluation Protocol=Ours2026.05 | 89.21 | 91.45 | 70.24 | 71.88 | |
| ROUTEGUARD-8BEvaluation Protocol=Ours2026.05 | 89.19 | 91.48 | 70.99 | 72.21 | |
| QWEN3GUARD-4BEvaluation Protocol=Fine-tuned2026.05 | 89.17 | 91.38 | 69.93 | 71.27 | |
| QWEN3GUARD-8BEvaluation Protocol=Fine-tuned2026.05 | 88.99 | 91.26 | 70.54 | 71.84 | |
| SHIELDGEMMA-9BEvaluation Protocol=Fine-tuned2026.05 | 88.58 | 90.88 | 69.66 | 71.09 | |
| LLAMAGUARD2-8BEvaluation Protocol=Fine-tuned2026.05 | 88.29 | 90.6 | 68.99 | 70.77 | |
| LLAMAGUARD3-8BEvaluation Protocol=Fine-tuned2026.05 | 88.2 | 90.46 | 68.8 | 70.35 | |
| SHIELDGEMMA-2BEvaluation Protocol=Fine-tuned2026.05 | 88.11 | 90.54 | 69.37 | 71.2 | |
| LLAMAGUARD3-1BEvaluation Protocol=Fine-tuned2026.05 | 87.6 | 89.94 | 67.63 | 70.01 | |
| QWEN3GUARD-0.6BEvaluation Protocol=Fine-tuned2026.05 | 87.41 | 89.81 | 62.65 | 69.83 | |
| ROUTEGUARD-0.6BEvaluation Protocol=Ours2026.05 | 86.28 | 88.67 | 67.76 | 70.58 | |
| GRAPH ROUTEREvaluation Protocol=Router Baseline2026.05 | 80.06 | 82.23 | 58.66 | 66.74 | |
| DC ROUTEREvaluation Protocol=Router Baseline2026.05 | 79.18 | 81.34 | 58.96 | 66.08 | |
| MLP ROUTEREvaluation Protocol=Router Baseline2026.05 | 78.45 | 80.39 | 57.84 | 65 | |
| QWEN3GUARD-4BEvaluation Protocol=0-shot2026.05 | 77 | 79.04 | 21.47 | 31.11 | |
| QWEN3GUARD-8BEvaluation Protocol=0-shot2026.05 | 76.14 | 80.23 | 26.63 | 37.5 | |
| KNN ROUTEREvaluation Protocol=Router Baseline2026.05 | 71.33 | 71.45 | 48.55 | 55.41 | |
| LLAMAGUARD3-8BEvaluation Protocol=0-shot2026.05 | 70 | 69.87 | 3.86 | 13.79 | |
| SHIELDGEMMA-2BEvaluation Protocol=0-shot2026.05 | 68.09 | 75.13 | 39.89 | 48.51 | |
| LLAMAGUARD2-8BEvaluation Protocol=0-shot2026.05 | 67.71 | 65.52 | 2.34 | 3.56 | |
| QWEN3GUARD-0.6BEvaluation Protocol=0-shot2026.05 | 66.5 | 68.83 | 10.31 | 23.4 | |
| SHIELDGEMMA-9BEvaluation Protocol=0-shot2026.05 | 63.74 | 75.61 | 40.79 | 48.38 | |
| SHIELDGEMMA-27BEvaluation Protocol=0-shot2026.05 | 52.07 | 55.3 | 16.72 | 20.42 | |
| LLAMAGUARD3-1BEvaluation Protocol=0-shot2026.05 | 45.13 | 25.2 | 1.17 | 3.55 |