List your company
ResearchFormerly WizWand

Safety Evaluation on JailBreak-R1

0LRM-JailBreak ScoreSTAR-1
-0.3441.9784.36.622May 9, 2026
Updated 5mo ago

Evaluation results

MethodLinks
STAR-1Model Backbone=DeepSeek-R1-Distill-Llama-8B, RL Training (GRPO)=true2026.050
SInternalModel Backbone=DeepSeek-R1-Distill-Llama-8B, RL Training (GRPO)=true2026.050
BaseModel Backbone=DeepSeek-R1-Distill-Qwen-14B, RL Training (GRPO)=true2026.050
SafeChainModel Backbone=DeepSeek-R1-Distill-Qwen-14B, RL Training (GRPO)=true2026.050
SInternalModel Backbone=DeepSeek-R1-Distill-Qwen-14B, RL Training (GRPO)=true2026.050.3
STAR-1Model Backbone=DeepSeek-R1-Distill-Qwen-14B, RL Training (GRPO)=true2026.050.6
SafeChainModel Backbone=DeepSeek-R1-Distill-Llama-8B, RL Training (GRPO)=true2026.051.6
SInternalModel Backbone=DeepSeek-R1-Distill-Qwen-7B, RL Training (GRPO)=true2026.052.2
BaseModel Backbone=DeepSeek-R1-Distill-Llama-8B, RL Training (GRPO)=true2026.052.9
STAR-1Model Backbone=DeepSeek-R1-Distill-Qwen-7B, RL Training (GRPO)=true2026.053.5
BaseModel Backbone=DeepSeek-R1-Distill-Qwen-7B, RL Training (GRPO)=true2026.058.3
SafeChainModel Backbone=DeepSeek-R1-Distill-Qwen-7B, RL Training (GRPO)=true2026.058.6