Prompt Moderation on Public Benchmarks Suite (ToxicChat, WildGuard, XSTest, OpenAI, Aegis2.0)
81.26ToxicChat ScoreQwen3Guard-8B-Gen (loose)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen3Guard-8B-Gen (loose)Model Size=8B, Evaluation Protocol=loose2026.02 | 81.26 | 86.43 | 89.07 | 80.97 | 82.39 | 84.02 | |
| BingoGuard-8BModel Size=8B2026.02 | 76.07 | 88.49 | 94.5 | 77.03 | 85.71 | 84.36 | |
| GuardReasoner-8BModel Size=8B2026.02 | 74.83 | 88.89 | 93.33 | 73.22 | 83.83 | 82.82 | |
| FlexGuard2026.02 | 74.4 | 87.5 | 92.34 | 81.14 | 85.44 | 84.02 | |
| WildGuard-7BModel Size=7B2026.02 | 65.93 | 88.45 | 94.55 | 72.1 | 80.99 | 80.4 | |
| Qwen3Guard-8B-Gen (strict)Model Size=8B, Evaluation Protocol=strict2026.02 | 63.84 | 89.11 | 90.82 | 68.54 | 86.23 | 79.71 | |
| LlamaGuard3-8BModel Size=8B2026.02 | 48.61 | 77.32 | 87.3 | 79.19 | 77.32 | 73.95 |