Text-based safety moderation on BeaverTails
87.3F1 ScoreRoblox Guard 1.0
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Roblox Guard 1.0Parameter Count=8B2025.12 | 87.3 | — | |
| Qwen3Guard-8B-strictAccess=Open-Source, Configuration=strict2026.06 | 86.69 | — | |
| Qwen3Guard-4B-Gen-strictStreaming=no, Moderation Policy=Gen-strict, Backbone Scale=4B2026.04 | 86.6 | — | |
| Qwen3Guard-8B-Gen-strictStreaming=no, Moderation Policy=Gen-strict, Backbone Scale=8B2026.04 | 86.6 | — | |
| Llama3-StreamGuard-8BStreaming=yes, Backbone Scale=8B2026.04 | 86.6 | — | |
| BingoGuardParameter Count=8B2025.12 | 86.4 | — | |
| Llama3-StreamGuard-3BStreaming=yes, Backbone Scale=3B2026.04 | 86.4 | — | |
| Llama3-StreamGuard-1BStreaming=yes, Backbone Scale=1B2026.04 | 86.2 | — | |
| SingGuard-2BAccess=Open-Source2026.06 | 86.01 | — | |
| Qwen3Guard-Stream-4B-strictStreaming=yes, Moderation Policy=strict, Backbone Scale=4B2026.04 | 86 | — | |
| Qwen3Guard-Stream-8B-strictStreaming=yes, Moderation Policy=strict, Backbone Scale=8B2026.04 | 85.9 | — | |
| SingGuard-8BAccess=Open-Source2026.06 | 85.85 | — | |
| SingGuard-4BAccess=Open-Source2026.06 | 85.82 | — | |
| YuFeng-XGuard-Reason-8BAccess=Open-Source2026.06 | 85.71 | — | |
| Qwen3Guard-Stream-8B-looseStreaming=yes, Moderation Policy=loose, Backbone Scale=8B2026.04 | 85.5 | — | |
| Qwen3Guard-Stream-4B-looseStreaming=yes, Moderation Policy=loose, Backbone Scale=4B2026.04 | 85.2 | — | |
| Qwen3Guard-8B-looseAccess=Open-Source, Configuration=loose2026.06 | 84.93 | — | |
| ShieldGemmaParameter Count=7B2025.12 | 84.8 | — | |
| Qwen3Guard-Stream-0.6B-strictStreaming=yes, Moderation Policy=strict, Backbone Scale=0.6B2026.04 | 84.5 | — | |
| GuardReasoner-VL-7BAccess=Open-Source2026.06 | 84.49 | — | |
| WildGuardParameter Count=7B2025.12 | 84.4 | — | |
| MD-JudgeSize=7B2026.03 | 84 | — | |
| Qwen3Guard-Stream-0.6B-looseStreaming=yes, Moderation Policy=loose, Backbone Scale=0.6B2026.04 | 84 | — | |
| OMNIGUARD-7BSize=7B2025.12 | 83.9 | 80.5 | |
| GPT-4o2025.12 | 83.8 | — | |
| Qwen2.5-72BSize=72B2025.12 | 83.6 | 80.5 | |
| GPT-4oSize=-2025.12 | 83.5 | 72.6 | |
| WildguardSize=7B2026.03 | 83 | — | |
| FanarGuardSize=4B2026.03 | 83 | — | |
| ThinkGuardSize=8B2025.12 | 82.7 | 81.6 | |
| Qwen3-VL-235BAccess=Open-Source2026.06 | 82.15 | — | |
| FanarGuardSize=4B2026.03 | 82 | — | |
| Qwen3-235BSize=235B2025.12 | 81.9 | 77.1 | |
| OMNIGUARD-3BSize=3B2025.12 | 81.8 | 82.6 | |
| GPT-5.1Access=Closed-Source2026.06 | 80.42 | — | |
| PolyGuard-MinistralSize=8B2026.03 | 80 | — | |
| PolyGuard-QwenSize=7B2026.03 | 80 | — | |
| Gemini3-ProAccess=Closed-Source2026.06 | 79.6 | — | |
| PolyGuard-MinistralSize=8B2026.03 | 79 | — | |
| PolyGuard-QwenSize=7B2026.03 | 78 | — | |
| WildGuardAccess=Open-Source2026.06 | 77.89 | — | |
| NemoGuardParameter Count=8B2025.12 | 77.6 | — | |
| LLaMA-3.3-70BSize=70B2025.12 | 76.8 | 71.3 | |
| ShieldGemma-2bSize=2B2026.03 | 76 | — | |
| Qwen2.5-7BSize=7B2025.12 | 75.3 | 72.6 | |
| GraniteGuardianAccess=Open-Source2026.06 | 72.91 | — | |
| LLaMA Guard 2Size=8B2025.12 | 72.3 | 73.5 | |
| LLaMA Guard 3Size=8B2025.12 | 71.2 | 73.5 | |
| PolyGuard-SmolSize=0.5B2026.03 | 71 | — | |
| PolyGuard-SmolSize=0.5B2026.03 | 71 | — | |
| ShieldGemma-2bSize=2B2026.03 | 71 | — | |
| Llama-Guard-3Size=8B2026.03 | 70 | — | |
| LlamaGuard3Parameter Count=8B2025.12 | 69.7 | — | |
| ShieldGemma-9BAccess=Open-Source2026.06 | 69.62 | — | |
| Llama Guard 3Access=Open-Source2026.06 | 67.78 | — | |
| Llama-Guard-3Size=8B2026.03 | 66 | — | |
| Qwen2.5-Omni-7BSize=7B2025.12 | 58.4 | 55.4 | |
| WildguardSize=7B2026.03 | 48 | — | |
| LLaMA Guard 1Size=7B2025.12 | 38.1 | 55.9 | |
| MD-JudgeSize=7B2026.03 | 31 | — |