List your company

Safety Evaluation on SafetyBench (accuracy)

89AccuracyInvThink SFT+RL
60.9268.2175.582.79Oct 2, 2025

Evaluation results

MethodLinks
InvThink SFT+RLBackbone=Qwen-3-8B2025.1089
InvThink SFTBackbone=Qwen-3-8B2025.1087
InvThinkBackbone=Qwen-3-8B2025.1085
General SFT+RLBackbone=Qwen-3-8B2025.1085
Gemini-2.5 ProBackbone=Teacher Model2025.1085
SafetyPromptBackbone=Qwen-3-8B2025.1084
CoTBackbone=Qwen-3-8B2025.1083
InvThink SFT+RLBackbone=Qwen-2.5-7B2025.1082
General SFTBackbone=Qwen-3-8B2025.1082
InvThink SFTBackbone=Qwen-2.5-7B2025.1079
InvThink SFT+RLBackbone=Gemma-7B-it2025.1077
General SFT+RLBackbone=Qwen-2.5-7B2025.1077
ToTBackbone=Qwen-3-8B2025.1077
InvThink SFTBackbone=Gemma-7B-it2025.1076
CoTBackbone=Qwen-2.5-7B2025.1076
InvThinkBackbone=Qwen-2.5-7B2025.1076
General SFTBackbone=Qwen-2.5-7B2025.1076
Zero-shotBackbone=Qwen-3-8B2025.1076
SafetyPromptBackbone=Qwen-2.5-7B2025.1075
General SFT+RLBackbone=Gemma-7B-it2025.1074
InvThinkBackbone=Gemma-7B-it2025.1073
Zero-shotBackbone=Qwen-2.5-7B2025.1073
Zero-shotBackbone=Gemma-7B-it2025.1072
General SFTBackbone=Gemma-7B-it2025.1072
ToTBackbone=Qwen-2.5-7B2025.1071
CoTBackbone=Gemma-7B-it2025.1069
SafetyPromptBackbone=Gemma-7B-it2025.1067
ToTBackbone=Gemma-7B-it2025.1062