Harmfulness Detection on Aegis
89.78Macro F1WildGuard
Evaluation results
| Method | Links | |
|---|---|---|
| WildGuardMethodology=Guard Model2025.02 | 89.78 | |
| MLPMBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 89.23 | |
| Ayub & MajumdarBackbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 88.11 | |
| GraniteGuardian-3-1-8BMethodology=Guard Model2025.02 | 87.78 | |
| MLPMBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 87.36 | |
| MLPMBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 85.13 | |
| Abdelnabi et al.Backbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 84.55 | |
| Abdelnabi et al.Backbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 84.47 | |
| Abdelnabi et al.Backbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 84.16 | |
| Abdelnabi et al.Backbone=OLMo2-7B-Inst, Methodology=Latent-Based2025.02 | 84.16 | |
| MLPMBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 83.49 | |
| SIRENBackbone=Qwen3-4B2026.04 | 82.9 | |
| SIRENBackbone=Llama3.1-8B2026.04 | 82.9 | |
| Ayub & MajumdarBackbone=Llama-8B-Inst, Methodology=Latent-Based2025.02 | 82.52 | |
| SIRENBackbone=Qwen3-0.6B2026.04 | 82.4 | |
| SIRENBackbone=Llama3.2-1B2026.04 | 82.1 | |
| Aegis-Guard-DMethodology=Guard Model2025.02 | 81 | |
| Ayub & MajumdarBackbone=Qwen3-8B-Inst, Methodology=Latent-Based2025.02 | 80 | |
| Ayub & MajumdarBackbone=Mistral-7B-Inst, Methodology=Latent-Based2025.02 | 79.6 | |
| Qwen3GuardBackbone=Qwen3-0.6B2026.04 | 78.8 | |
| Qwen3GuardBackbone=Qwen3-4B2026.04 | 78.2 | |
| ShieldGemma-9BMethodology=Guard Model2025.02 | 77.44 | |
| LlamaGuard3Methodology=Guard Model2025.02 | 71.74 | |
| LlamaGuard3Backbone=Llama3.1-8B2026.04 | 67.1 | |
| LlamaGuard3Backbone=Llama3.2-1B2026.04 | 59.5 |