Prompt Classification on Aegis 2.0
89.2F1 ScoreHaloGuard 1.0
Evaluation Results
| Method | Links | |
|---|---|---|
| HaloGuard 1.0Size=4B2026.07 | 89.2 | |
| HaloGuard 1.0Size=0.8B2026.07 | 87.9 | |
| NemotronReasoningModel Size=4B2026.01 | 87.3 | |
| YuFeng-XGuardModel Size=0.6B2026.01 | 87.1 | |
| NemoGuardSize=8B2026.07 | 86.8 | |
| PolyGuardModel Size=7B2026.01 | 86.6 | |
| Qwen3Guard-GenModel Size=8B, Evaluation protocol=strict2026.01 | 86.6 | |
| YuFeng-XGuardModel Size=8B2026.01 | 86.4 | |
| Qwen3Guard-GenModel Size=4B, Evaluation protocol=strict2026.01 | 86.3 | |
| NemotronGuardV2Model Size=8B2026.01 | 86.3 | |
| NemoGuard-8BSafeguard Type=Post-hoc2026.02 | 86.3 | |
| PolyGuard-QwenSize=7B2026.07 | 86.3 | |
| Qwen3Guard-GenSize=8B2026.07 | 86.1 | |
| Qwen3Guard-GenSize=4B2026.07 | 85.8 | |
| Qwen3Guard-GenModel Size=0.6B, Evaluation protocol=strict2026.01 | 85.2 | |
| Qwen3Guard-GenSize=0.6B2026.07 | 85 | |
| NExT-Guard*Safeguard Type=Streaming2026.02 | 84.8 | |
| NExT-GuardSafeguard Type=Streaming2026.02 | 84 | |
| Qwen3Guard-GenModel Size=0.6B, Evaluation protocol=loose2026.01 | 83.2 | |
| Qwen3Guard-GenModel Size=8B, Evaluation protocol=loose2026.01 | 82.6 | |
| Qwen3Guard-GenModel Size=4B, Evaluation protocol=loose2026.01 | 82.5 | |
| Qwen3Guard-0.6B-GenSafeguard Type=Post-hoc2026.02 | 82.4 | |
| GPT-OSS-SafeGuardModel Size=20B2026.01 | 82.2 | |
| WildGuardModel Size=7B2026.01 | 81.5 | |
| Qwen3Guard-8B-GenSafeguard Type=Post-hoc2026.02 | 81.3 | |
| Qwen3Guard-4B-GenSafeguard Type=Post-hoc2026.02 | 81.2 | |
| Qwen3Guard-0.6B-StreamSafeguard Type=Streaming2026.02 | 81.1 | |
| WildGuardSize=7B2026.07 | 80.7 | |
| ShieldGemmaModel Size=9B2026.01 | 79.9 | |
| WildGuard-7BSafeguard Type=Post-hoc2026.02 | 79.8 | |
| Qwen3Guard-8B-StreamSafeguard Type=Streaming2026.02 | 79.3 | |
| Qwen3Guard-4B-StreamSafeguard Type=Streaming2026.02 | 79.1 | |
| SCM-7BSafeguard Type=Streaming2026.02 | 78.6 | |
| KelpSafeguard Type=Streaming2026.02 | 78.2 | |
| Llama3GuardModel Size=8B2026.01 | 77.2 | |
| LlamaGuard3-8BSafeguard Type=Post-hoc2026.02 | 75.7 | |
| ShieldGemma-9BSafeguard Type=Post-hoc2026.02 | 71.8 | |
| ShieldGemmaSize=27B2026.07 | 71.6 | |
| Llama4GuardModel Size=12B2026.01 | 71.5 | |
| LlamaGuard4Size=12B2026.07 | 70.6 | |
| ShieldGemma-27BSafeguard Type=Post-hoc2026.02 | 70.3 | |
| LlamaGuard4-12BSafeguard Type=Post-hoc2026.02 | 70.1 |