Prompt classification on SimpST
100F1 ScorePolyGuard
Evaluation Results
| Method | Links | |
|---|---|---|
| PolyGuardModel Size=7B2026.01 | 100 | |
| YuFeng-XGuardModel Size=8B2026.01 | 100 | |
| YuFeng-XGuard-Reason-8BParams=8B2026.07 | 100 | |
| Stage1-SFT-v0Params=4B2026.07 | 100 | |
| Stage1-SFT-v1Params=4B2026.07 | 100 | |
| Stage1-SFT-v2Params=4B2026.07 | 100 | |
| Stage1-SFT-v3Params=4B2026.07 | 100 | |
| Stage1-SFT-v4Params=4B2026.07 | 100 | |
| Stage2-SFTParams=4B2026.07 | 100 | |
| Stage3-DPOParams=4B2026.07 | 100 | |
| NemotronReasoningModel Size=4B2026.01 | 99.5 | |
| Qwen3Guard-GenModel Size=4B, Evaluation protocol=strict2026.01 | 99.5 | |
| YuFeng-XGuardModel Size=0.6B2026.01 | 99.5 | |
| Llama3GuardModel Size=8B2026.01 | 99.5 | |
| WildGuardModel Size=7B2026.01 | 99.5 | |
| GPT-OSS-SafeGuardModel Size=20B2026.01 | 99.5 | |
| Qwen3Guard-GenModel Size=8B, Evaluation protocol=strict2026.01 | 99.5 | |
| NExT-GuardSafeguard Type=Streaming2026.02 | 99.5 | |
| Qwen3Guard-4B-GenParams=4B2026.07 | 99.5 | |
| Qwen3Guard-8B-GenParams=8B2026.07 | 99.5 | |
| YuFeng-XGuard-Reason-0.6BParams=0.6B2026.07 | 99.5 | |
| Qwen3Guard-0.8B-GenParams=0.8B2026.07 | 99 | |
| LlamaGuard3-8BSafeguard Type=Post-hoc2026.02 | 98.8 | |
| Qwen3Guard-GenModel Size=0.6B, Evaluation protocol=strict2026.01 | 98.5 | |
| Llama4GuardModel Size=12B2026.01 | 98.5 | |
| NemotronGuardV2Model Size=8B2026.01 | 98.5 | |
| WildGuard-7BSafeguard Type=Post-hoc2026.02 | 98.2 | |
| NemoGuard-8BSafeguard Type=Post-hoc2026.02 | 98 | |
| NExT-Guard*Safeguard Type=Streaming2026.02 | 98 | |
| Qwen3Guard-4B-StreamSafeguard Type=Streaming2026.02 | 97.9 | |
| Qwen3Guard-8B-StreamSafeguard Type=Streaming2026.02 | 97.7 | |
| Qwen3Guard-GenModel Size=4B, Evaluation protocol=loose2026.01 | 97.4 | |
| Qwen3Guard-GenModel Size=8B, Evaluation protocol=loose2026.01 | 97.4 | |
| Qwen3Guard-8B-GenSafeguard Type=Post-hoc2026.02 | 96.9 | |
| LlamaGuard4-12BSafeguard Type=Post-hoc2026.02 | 96.8 | |
| KelpSafeguard Type=Streaming2026.02 | 96.7 | |
| Qwen3Guard-4B-GenSafeguard Type=Post-hoc2026.02 | 96.5 | |
| SCM-7BSafeguard Type=Streaming2026.02 | 96.5 | |
| Qwen3Guard-0.6B-StreamSafeguard Type=Streaming2026.02 | 95.9 | |
| Qwen3Guard-GenModel Size=0.6B, Evaluation protocol=loose2026.01 | 95.8 | |
| ShieldGemmaModel Size=9B2026.01 | 95.7 | |
| Qwen3Guard-0.6B-GenSafeguard Type=Post-hoc2026.02 | 94.5 | |
| ShieldGemma-27BSafeguard Type=Post-hoc2026.02 | 83.9 | |
| ShieldGemma-9BSafeguard Type=Post-hoc2026.02 | 83.1 |