List your company
ResearchFormerly WizWand

Safety Classification on ASSEBench

92.04AccuracyTRACE
57.35666.360575.36584.3695May 30, 2026
Updated 5mo ago

Evaluation results

MethodLinks
TRACEBackbone=Qwen3Guard-Gen-4B2026.0592.0490.5588.77
TRACEBackbone=Qwen3-8B2026.0591.5789.7587.19
TRACEBackbone=Qwen3-4B-Instruct-25072026.0591.3888.9886.04
TRACEBackbone=Llama-3.1-8B2026.0589.7286.9184.62
MAGEBackbone=Qwen3-4B-Instruct-25072026.0586.4779.3372.18
SFTBackbone=Qwen3-4B-Instruct-25072026.0584.2277.0668.9
MAGEBackbone=Qwen3Guard-Gen-4B2026.0583.4776.1370.28
MAGEBackbone=Qwen3-8B2026.0583.1483.7680.93
SFTBackbone=Qwen3Guard-Gen-4B2026.0581.0974.1568.34
AABackbone=Qwen3-8B2026.0580.8281.8479.33
MAGEBackbone=Llama-3.1-8B2026.0580.6381.1272.77
SFTBackbone=Qwen3-8B2026.0580.1772.2764.45
AABackbone=Llama-3.1-8B2026.0577.9979.4170.67
SFTBackbone=Llama-3.1-8B2026.0576.5663.4157.17
AABackbone=Qwen3Guard-Gen-4B2026.0574.1972.4467.01
AABackbone=Qwen3-4B-Instruct-25072026.0571.0658.9260.3
BaseBackbone=Qwen3-4B-Instruct-25072026.0563.2344.5741.09
BaseBackbone=Qwen3Guard-Gen-4B2026.0562.6434.3623.66
BaseBackbone=Llama-3.1-8B2026.0561.5525.6916.08
BaseBackbone=Qwen3-8B2026.0558.6949.8749.85