Harmfulness Detection on WildGuard
90.6Macro F1 ScoreQwen3Guard
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3GuardBackbone=Qwen3-4B2026.04 | 90.6 | |
| ConsisGuardCategory=Reasoning-based Guardrails, Model Size=7B2026.05 | 89.96 | |
| ConsisGuardCategory=Reasoning-based Guardrails, Model Size=3B2026.05 | 89.8 | |
| GuardReasonerCategory=Reasoning-based Guardrails, Model Size=8B2026.05 | 89.17 | |
| Qwen3GuardBackbone=Qwen3-0.6B2026.04 | 89.1 | |
| GuardReasonerCategory=Reasoning-based Guardrails, Model Size=3B2026.05 | 89.01 | |
| WildGuardCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 88.9 | |
| SIRENBackbone=Qwen3-4B2026.04 | 88.3 | |
| SIRENBackbone=Llama3.1-8B2026.04 | 86.7 | |
| SIRENBackbone=Qwen3-0.6B2026.04 | 86.5 | |
| SIRENBackbone=Llama3.2-1B2026.04 | 86.5 | |
| Gemini 1.5+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 82.89 | |
| LlamaGuard3Backbone=Llama3.1-8B2026.04 | 81.3 | |
| qwen3-CoTCategory=CoT-Based LRM, Model Size=32B2026.05 | 80.24 | |
| GPT-4o+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 80.13 | |
| GPT-4+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 80.05 | |
| o1-pre+CoTCategory=CoT-Based LRM, Model Size=-2026.05 | 79.82 | |
| GuardReasonerCategory=Reasoning-based Guardrails, Model Size=3B2026.05 | 79.7 | |
| ConsisGuardCategory=Reasoning-based Guardrails, Model Size=7B2026.05 | 78.71 | |
| LlamaGuard3Backbone=Llama3.2-1B2026.04 | 78.6 | |
| Aegis Guard DefCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 78.5 | |
| ConsisGuardCategory=Reasoning-based Guardrails, Model Size=3B2026.05 | 78.39 | |
| GuardReasonerCategory=Reasoning-based Guardrails, Model Size=8B2026.05 | 78.2 | |
| MD-JudgeCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 76.8 | |
| QWQ+CoTCategory=CoT-Based LRM, Model Size=32B2026.05 | 76.37 | |
| LLaMA Guard 3Category=Discriminant-based Guardrails, Model Size=8B2026.05 | 76.18 | |
| Gemini 1.5+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 75.93 | |
| WildGuardCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 75.4 | |
| qwen3-CoTCategory=CoT-Based LRM, Model Size=32B2026.05 | 74.48 | |
| GPT-4o+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 71.69 | |
| Aegis Guard PerCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 71.5 | |
| LLaMA Guard 2Category=Discriminant-based Guardrails, Model Size=8B2026.05 | 70.9 | |
| LLaMA Guard 3Category=Discriminant-based Guardrails, Model Size=8B2026.05 | 70.8 | |
| o1-pre+CoTCategory=CoT-Based LRM, Model Size=-2026.05 | 68.26 | |
| QWQ+CoTCategory=CoT-Based LRM, Model Size=32B2026.05 | 67.44 | |
| GPT-4+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 66.68 | |
| LLaMA Guard 2Category=Discriminant-based Guardrails, Model Size=8B2026.05 | 66.5 | |
| Claude 3.5+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 65.48 | |
| ShieldGemmaCategory=Discriminant-based Guardrails, Model Size=9B2026.05 | 57.74 | |
| Aegis Guard PerCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 56.4 | |
| Aegis Guard DefCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 49.1 | |
| ShieldGemmaCategory=Discriminant-based Guardrails, Model Size=9B2026.05 | 47 | |
| Claude 3.5+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 32.36 | |
| ShieldGemmaCategory=Discriminant-based Guardrails, Model Size=2B2026.05 | 20.13 | |
| ModerationCategory=Discriminant-based Guardrails, Model Size=-2026.05 | 16.9 | |
| ModerationCategory=Discriminant-based Guardrails, Model Size=-2026.05 | 12.1 | |
| ShieldGemmaCategory=Discriminant-based Guardrails, Model Size=2B2026.05 | 9.36 |