Harmful Content Evaluation on Challenging Harmful Content Evaluation Dataset
84.08AccuracyQwen3Guard 4B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3Guard 4BModel Parameters=4B2025.12 | 84.08 | 5.31 | 9.5 | |
| Qwen3GuardParameters=4B2026.03 | 84.08 | 5.31 | 9.5 | |
| Qwen3Guard 8BModel Parameters=8B2025.12 | 81.84 | 7.82 | 9.22 | |
| Qwen3GuardParameters=8B2026.03 | 81.84 | 7.82 | 9.22 | |
| Qwen3Guard 0.6BModel Parameters=0.6B2025.12 | 77.65 | 6.98 | 14.25 | |
| Qwen3GuardParameters=0.6B2026.03 | 77.65 | 6.98 | 14.25 | |
| DeepSeek2025.12 | 70.67 | 8.66 | 19.55 | |
| DeepSeek2026.03 | 70.67 | 8.66 | 19.55 | |
| GPT-52025.12 | 70.39 | 7.54 | 20.39 | |
| GPT-42026.03 | 70.39 | 7.54 | 20.39 | |
| Keyword-based Detection2025.12 | 60.89 | 14.25 | 23.74 | |
| Keyword-based2026.03 | 60.89 | 14.25 | 23.74 | |
| Llama-Prompt-Guard-2-22MModel Parameters=22M2025.12 | 49.16 | 12.29 | 37.43 | |
| Llama-Prompt-Guard-2Parameters=22M2026.03 | 49.16 | 12.29 | 37.43 | |
| Llama-Prompt-Guard-2-86MModel Parameters=86M2025.12 | 41.34 | 34.07 | 23.46 | |
| Llama-Prompt-Guard-2Parameters=86M2026.03 | 41.34 | 34.07 | 23.46 |