Safety Classification on SafeRLHF
0.94F1 ScoreQwen3Guard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3GuardSize=8B, Mode=loose, Reasoning=false2025.12 | 0.94 | — | |
| AprielGuardSize=8B, Reasoning=true2025.12 | 0.93 | — | |
| IBM Granite Guardian 3.2Size=5B, Reasoning=false2025.12 | 0.92 | — | |
| AprielGuardSize=8B, Reasoning=false2025.12 | 0.92 | — | |
| IBM Granite Guardian 3.3Size=8B, Reasoning=false2025.12 | 0.91 | — | |
| Qwen3GuardSize=8B, Mode=strict, Reasoning=false2025.12 | 0.91 | — | |
| gpt-oss-safeguardSize=20B, Reasoning=true2025.12 | 0.91 | — | |
| IBM Granite Guardian 3.1Size=2B, Reasoning=false2025.12 | 0.9 | — | |
| IBM Granite Guardian 3.3Size=8B, Reasoning=true2025.12 | 0.9 | — | |
| Llama Guard 3Reasoning=false2025.12 | 0.89 | — | |
| NExT-GuardSafeguard Type=Streaming2026.02 | 0.888 | — | |
| Llama Guard 2Reasoning=false2025.12 | 0.88 | — | |
| Llama Guard 4Reasoning=false2025.12 | 0.88 | — | |
| IBM Granite Guardian 3.2Size=3B, Reasoning=false2025.12 | 0.88 | — | |
| NExT-Guard*Safeguard Type=Streaming2026.02 | 0.833 | — | |
| BeaverDam 7BModel Category=LLM Guard Model2026.02 | 0.721 | — | |
| Qwen3Guard Gen 8BModel Category=LLM Guard Model, Detection Mode=strict2026.02 | 0.7054 | — | |
| GuardReasoner 8BModel Category=LLM Guard Model2026.02 | 0.7004 | — | |
| GuardReasoner-Omni 4BModel Category=VLM Guard Model2026.02 | 0.6844 | — | |
| GuardReasoner-Omni 2BModel Category=VLM Guard Model2026.02 | 0.6814 | — | |
| GuardReasoner-VL 7BModel Category=VLM Guard Model2026.02 | 0.6663 | — | |
| GuardReasoner-VL 3BModel Category=VLM Guard Model2026.02 | 0.6634 | — | |
| MD-Judge 7BModel Category=LLM Guard Model2026.02 | 0.647 | — | |
| WildGuard 7BModel Category=LLM Guard Model2026.02 | 0.642 | — | |
| Qwen3Guard-4B-GenSafeguard Type=Post-hoc2026.02 | 0.637 | — | |
| Qwen3Guard-4B-StreamSafeguard Type=Streaming2026.02 | 0.637 | — | |
| Qwen3Guard Gen 8BModel Category=LLM Guard Model, Detection Mode=loose2026.02 | 0.6348 | — | |
| PolyGuard-Qwen 7BModel Category=LLM Guard Model2026.02 | 0.6344 | — | |
| Qwen3Guard-0.6B-GenSafeguard Type=Post-hoc2026.02 | 0.634 | — | |
| Qwen3Guard-8B-GenSafeguard Type=Post-hoc2026.02 | 0.632 | — | |
| WildGuard-7BSafeguard Type=Post-hoc2026.02 | 0.629 | — | |
| Qwen3Guard-8B-StreamSafeguard Type=Streaming2026.02 | 0.617 | — | |
| Qwen3Guard-0.6B-StreamSafeguard Type=Streaming2026.02 | 0.611 | — | |
| HarmBench LLaMA 13BModel Category=LLM Guard Model2026.02 | 0.6 | — | |
| SCM-7BSafeguard Type=Streaming2026.02 | 0.598 | — | |
| Aegis Guard Defensive 7BModel Category=LLM Guard Model2026.02 | 0.593 | — | |
| NemoGuard-8BSafeguard Type=Post-hoc2026.02 | 0.565 | — | |
| ShieldGemmaSize=9B, Reasoning=false2025.12 | 0.56 | — | |
| Aegis Guard Permissive 7BModel Category=LLM Guard Model2026.02 | 0.559 | — | |
| KelpSafeguard Type=Streaming2026.02 | 0.54 | — | |
| HarmBench Mistral 7BModel Category=LLM Guard Model2026.02 | 0.524 | — | |
| ShieldGemma-27BSafeguard Type=Post-hoc2026.02 | 0.511 | — | |
| ShieldGemma 9BModel Category=LLM Guard Model2026.02 | 0.4707 | — | |
| LLaMA Guard 3 8BModel Category=LLM Guard Model2026.02 | 0.4436 | — | |
| LlamaGuard3-8BSafeguard Type=Post-hoc2026.02 | 0.441 | — | |
| LLaMA Guard 4 12BModel Category=VLM Guard Model2026.02 | 0.4331 | — | |
| ShieldGemma-9BSafeguard Type=Post-hoc2026.02 | 0.431 | — | |
| LlamaGuard4-12BSafeguard Type=Post-hoc2026.02 | 0.415 | — | |
| LlamaGuard3Backbone=Llama3.2-1B2026.04 | — | 83.3 | |
| LlamaGuard3Backbone=Llama3.1-8B2026.04 | — | 86.2 | |
| Qwen3GuardBackbone=Qwen3-0.6B2026.04 | — | 86.9 | |
| Qwen3GuardBackbone=Qwen3-4B2026.04 | — | 89.2 | |
| SIRENBackbone=Qwen3-0.6B2026.04 | — | 91.6 | |
| SIRENBackbone=Llama3.2-1B2026.04 | — | 92 | |
| SIRENBackbone=Qwen3-4B2026.04 | — | 93.2 | |
| SIRENBackbone=Llama3.1-8B2026.04 | — | 92.5 |