List your company
ResearchFormerly WizWand

Harmfulness Evaluation on AdvBench

1.04Harmfulness ScoreBase
0.8821.94853.0154.0815Aug 6, 2025Aug 18, 2025Aug 30, 2025Sep 11, 2025Sep 23, 2025Oct 5, 2025Oct 17, 2025
Updated 4mo ago

Evaluation results

MethodLinks
Base2025.101.04
Base2025.101.04
SAFEPATH-FTModel=R1-Llama-8B2025.081.06
ThinkingIModel=R1-Qwen-7B2025.081.08
ThinkingIModel=R1-Llama-8B2025.081.14
SAFEPATH-FTModel=R1-Qwen-7B2025.081.16
SafeKeyModel=R1-Llama-8B2025.081.34
ReasoningGuardModel=R1-Llama-8B2025.081.4
RealSafe-R1Model=R1-Qwen-7B2025.081.42
SafeKeyModel=R1-Qwen-7B2025.081.42
SAFEPATH-ZSModel=R1-Llama-8B2025.081.48
RealSafe-R1Model=R1-Llama-8B2025.081.52
ReasoningGuardModel=R1-Qwen-7B2025.081.78
SAFEPATH-ZSModel=R1-Qwen-7B2025.082.1
Self-ReminderModel=R1-Qwen-7B2025.082.32
Self-ReminderModel=R1-Llama-8B2025.082.32
ParaphraseModel=R1-Llama-8B2025.082.92
SmoothLLMModel=R1-Llama-8B2025.083.1
ParaphraseModel=R1-Qwen-7B2025.083.34
No DefenseModel=R1-Llama-8B2025.083.54
SmoothLLMModel=R1-Qwen-7B2025.083.58
SafeDecodingModel=R1-Llama-8B2025.083.64
No DefenseModel=R1-Qwen-7B2025.083.74
SafeDecodingModel=R1-Qwen-7B2025.083.74
SFT2025.104.87
SFT2025.104.87
RLVR2025.104.99
RLVR2025.104.99