Response Harmfulness Detection on BeaverTails
89.9F1 ScoreBeaverDam 7B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BeaverDam 7BModel Category=LLM Guard Model2026.02 | 89.9 | — | |
| BeaverDamModel Size=7B2026.05 | 89.9 | — | |
| ConsisGuardCategory=Reasoning-based Guardrails, Model Size=7B2026.05 | 88.3 | — | |
| GuardReasoner 8BModel Category=LLM Guard Model2026.02 | 87.6 | — | |
| GuardReasonerModel Size=8B2026.05 | 87.6 | — | |
| GuardReasonerCategory=Reasoning-based Guardrails, Model Size=8B2026.05 | 87.6 | — | |
| ConsisGuardCategory=Reasoning-based Guardrails, Model Size=3B2026.05 | 87.44 | — | |
| GuardReasonerModel Size=3B2026.05 | 86.72 | — | |
| GuardReasonerCategory=Reasoning-based Guardrails, Model Size=3B2026.05 | 86.72 | — | |
| MD-Judge 7BModel Category=LLM Guard Model2026.02 | 86.7 | — | |
| MD-JudgeModel Size=7B2026.05 | 86.7 | — | |
| MD-JudgeCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 86.7 | — | |
| COLAGUARDModel Size=8B2026.05 | 86.55 | — | |
| Qwen3Guard Gen 8BModel Category=LLM Guard Model, Detection Mode=strict2026.02 | 86.48 | — | |
| COLAGUARDModel Size=3B2026.05 | 86.29 | — | |
| GuardReasoner-Omni 4BModel Category=VLM Guard Model2026.02 | 86.04 | — | |
| GuardReasoner-Omni 2BModel Category=VLM Guard Model2026.02 | 85.89 | — | |
| GuardReasonerModel Size=1B2026.05 | 85.84 | — | |
| Qwen3Guard Gen 8BModel Category=LLM Guard Model, Detection Mode=loose2026.02 | 85.35 | — | |
| GuardReasoner-VL 3BModel Category=VLM Guard Model2026.02 | 85.19 | — | |
| GuardReasoner-VL 7BModel Category=VLM Guard Model2026.02 | 84.99 | — | |
| WildGuardCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 84.66 | — | |
| WildGuard 7BModel Category=LLM Guard Model2026.02 | 84.4 | — | |
| WildGuardModel Size=7B2026.05 | 84.4 | — | |
| GPT-4o+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 83.41 | — | |
| Gemini 1.5+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 83.27 | — | |
| Claude 3.5+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 83.2 | — | |
| GPT-4o+CoTModel Size=Unknown2026.05 | 82.26 | — | |
| o1-pre+CoTCategory=CoT-Based LRM, Model Size=-2026.05 | 80.89 | — | |
| GPT-4+CoTCategory=CoT-Based LLM, Model Size=-2026.05 | 80.44 | — | |
| o1-previewModel Size=Unknown2026.05 | 79.96 | — | |
| PolyGuard-Qwen 7BModel Category=LLM Guard Model2026.02 | 79.39 | — | |
| qwen3-CoTCategory=CoT-Based LRM, Model Size=32B2026.05 | 78.7 | — | |
| GPT-4oModel Size=Unknown2026.05 | 78.63 | — | |
| QWQ+CoTCategory=CoT-Based LRM, Model Size=32B2026.05 | 77.89 | — | |
| QwQ-previewModel Size=32B2026.05 | 77.26 | — | |
| HarmBench LLaMA 13BModel Category=LLM Guard Model2026.02 | 77.1 | — | |
| HarmBench LLaMAModel Size=13B2026.05 | 77.1 | — | |
| HarmBench Mistral 7BModel Category=LLM Guard Model2026.02 | 75.2 | — | |
| HarmBench MistralModel Size=7B2026.05 | 75.2 | — | |
| Aegis Guard Defensive 7BModel Category=LLM Guard Model2026.02 | 74.7 | — | |
| Aegis Guard DefensiveModel Size=7B2026.05 | 74.7 | — | |
| Aegis Guard DefCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 74.7 | — | |
| Aegis Guard Permissive 7BModel Category=LLM Guard Model2026.02 | 73.8 | — | |
| Aegis Guard PermissiveModel Size=7B2026.05 | 73.8 | — | |
| Aegis Guard PerCategory=Discriminant-based Guardrails, Model Size=7B2026.05 | 73.8 | — | |
| LLaMA Guard 2Model Size=8B2026.05 | 71.8 | — | |
| LLaMA Guard 2Category=Discriminant-based Guardrails, Model Size=8B2026.05 | 71.8 | — | |
| LLaMA Guard 4 12BModel Category=VLM Guard Model2026.02 | 69.51 | — | |
| LLaMA Guard 3 8BModel Category=LLM Guard Model2026.02 | 67.84 | — | |
| LLaMA Guard 3Model Size=8B2026.05 | 67.84 | — | |
| LLaMA Guard 3Category=Discriminant-based Guardrails, Model Size=8B2026.05 | 67.84 | — | |
| LLaMA GuardModel Size=7B2026.05 | 67.1 | — | |
| ShieldGemma 9BModel Category=LLM Guard Model2026.02 | 63.61 | — | |
| ShieldGemmaModel Size=9B2026.05 | 63.61 | — | |
| ShieldGemmaCategory=Discriminant-based Guardrails, Model Size=9B2026.05 | 63.61 | — | |
| ShieldGemmaModel Size=2B2026.05 | 30.97 | — | |
| ShieldGemmaCategory=Discriminant-based Guardrails, Model Size=2B2026.05 | 30.97 | — | |
| ModerationCategory=Discriminant-based Guardrails, Model Size=-2026.05 | 15.7 | — | |
| LlamaGuard3Backbone=Llama3.2-1B2026.04 | — | 70 | |
| LlamaGuard3Backbone=Llama3.1-8B2026.04 | — | 68.8 | |
| Qwen3GuardBackbone=Qwen3-0.6B2026.04 | — | 77.1 | |
| Qwen3GuardBackbone=Qwen3-4B2026.04 | — | 80.1 | |
| SIRENBackbone=Qwen3-0.6B2026.04 | — | 83.5 | |
| SIRENBackbone=Llama3.2-1B2026.04 | — | 83.7 | |
| SIRENBackbone=Qwen3-4B2026.04 | — | 84.3 | |
| SIRENBackbone=Llama3.1-8B2026.04 | — | 83.8 |