Streaming Safety Detection on WildGuard (test)
83.45Det@1SentGuard
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SentGuard2026.06 | 83.45 | 88.73 | 92.61 | 94.37 | 2.58 | 4.49 | |
| WildGuard-7B2026.06 | 65.49 | 71.48 | 76.76 | 78.52 | 9.24 | 10.95 | |
| Qwen3Guard-8B-Genmode=generation2026.06 | 64.08 | 71.48 | 77.11 | 79.58 | 8.25 | 7.93 | |
| LlamaGuard3-8B2026.06 | 59.86 | 63.03 | 66.55 | 69.72 | 11.34 | 14.81 | |
| LlamaGuard4-12B2026.06 | 36.62 | 41.9 | 53.52 | 59.86 | 15.77 | 7.65 | |
| Qwen3Guard-8B-Streammode=streaming2026.06 | 33.8 | 52.82 | 64.08 | 67.25 | 13.18 | 4 | |
| GPT-5.52026.06 | 23.36 | 39.42 | 61.31 | 72.63 | 8.22 | 16.33 | |
| Gemini-3.5-Flash2026.06 | 18.38 | 29.41 | 40.44 | 47.79 | 18.22 | 6.47 |