Text-based safety moderation on Toxic Chat
82F1 ScoreGGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GGuardType=Prompt2025.12 | 82 | — | |
| Roblox Guard 1.0Parameter Count=8B2025.12 | 79.1 | — | |
| ShieldGemmaType=Prompt2025.12 | 78.4 | — | |
| BingoGuardParameter Count=8B2025.12 | 75.7 | — | |
| WildGuardType=Prompt2025.12 | 70.8 | — | |
| WildGuardParameter Count=7B2025.12 | 70.8 | — | |
| ShieldGemmaParameter Count=7B2025.12 | 70.2 | — | |
| Llama GuardType=Prompt2025.12 | 70 | — | |
| GPT-4o2025.12 | 68.1 | — | |
| OMNIGUARD-3BSize=3B2025.12 | 67 | 95.6 | |
| Qwen3-235BSize=235B2025.12 | 66 | 95.5 | |
| Qwen2.5-Omni-7BSize=7B2025.12 | 65.2 | 93.9 | |
| OAI ModeratorType=Prompt2025.12 | 61 | — | |
| NemoGuardParameter Count=8B2025.12 | 59.5 | — | |
| Qwen2.5-7BSize=7B2025.12 | 58.3 | 95.1 | |
| OMNIGUARD-7BSize=7B2025.12 | 58.2 | 95.3 | |
| Qwen2.5-72BSize=72B2025.12 | 54.4 | 94.3 | |
| LLaMA-3.3-70BSize=70B2025.12 | 53.8 | 94.3 | |
| GPT-4oSize=-2025.12 | 51.2 | 94.8 | |
| LlamaGuard3Parameter Count=8B2025.12 | 50.9 | — | |
| ThinkGuardSize=8B2025.12 | 49.8 | 92.8 | |
| LLaMA Guard 3Size=8B2025.12 | 37.7 | 93.3 | |
| LLaMA Guard 2Size=8B2025.12 | 30.9 | 93.7 | |
| LLaMA Guard 1Size=7B2025.12 | 23.3 | 92.9 |