List your company

Safety classification on Pre-Ex-Bench (test)

92.96AccuracyTRACE
56.40465.894575.38584.8755May 30, 2026

Evaluation results

MethodLinks
TRACEBackbone=Qwen3-4B-Instruct-25072026.0592.9691.7994.0688.53
TRACEBackbone=Qwen2.5-7B-Instruct2026.0590.6288.359185.85
SFTBackbone=Qwen3-4B-Instruct-25072026.0585.558093.6769.81
MAGEBackbone=Qwen3-4B-Instruct-25072026.0580.0871.8286.6761.32
TRACEBackbone=Qwen2.5-3B-Instruct2026.0570.763.0565.9860.38
MAGEBackbone=Qwen2.5-3B-Instruct2026.0563.2826.5677.2716.04
BaseBackbone=Qwen2.5-7B-Instruct2026.0563.2130.3658.6220.48
MAGEBackbone=Qwen2.5-7B-Instruct2026.0561.9120.636512.26
BaseBackbone=Qwen2.5-3B-Instruct2026.0560.1419.9151.2212.35
SFTBackbone=Qwen2.5-3B-Instruct2026.0558.597.02503.77
BaseBackbone=Qwen3-4B-Instruct-25072026.0558.2513.66707.57
SFTBackbone=Qwen2.5-7B-Instruct2026.0557.811.82250.94