List your company

Safety Classification on Pre-Ex-Bench

94.01AccuracyTRACE
58.327667.591376.85586.1187May 30, 2026

Evaluation results

MethodLinks
TRACEBackbone=Llama-3.1-8B2026.0594.0192.1388.79
TRACEBackbone=Qwen3-4B-Instruct-25072026.0593.8892.2391.13
TRACEBackbone=Qwen3Guard-Gen-4B2026.0593.6291.5588.59
TRACEBackbone=Qwen3-8B2026.0592.0689.6989.27
MAGEBackbone=Qwen3-8B2026.0590.8788.3585.21
SFTBackbone=Qwen3-8B2026.0590.4987.2983.06
MAGEBackbone=Qwen3-4B-Instruct-25072026.0588.5285.7466.09
SFTBackbone=Qwen3-4B-Instruct-25072026.0586.9183.8181.07
MAGEBackbone=Llama-3.1-8B2026.0581.8972.3471.28
MAGEBackbone=Qwen3Guard-Gen-4B2026.0581.0557.8252.41
SFTBackbone=Llama-3.1-8B2026.0579.3964.9656.99
SFTBackbone=Qwen3Guard-Gen-4B2026.0579.0356.3250.38
AABackbone=Llama-3.1-8B2026.0570.4268.4267.25
AABackbone=Qwen3-8B2026.0569.8440.7929.25
BaseBackbone=Llama-3.1-8B2026.0562.216.8410.54
AABackbone=Qwen3-4B-Instruct-25072026.0560.5653.8745.45
BaseBackbone=Qwen3-8B2026.0560.5315.4412.83
BaseBackbone=Qwen3Guard-Gen-4B2026.0560.4710.5113.43
AABackbone=Qwen3Guard-Gen-4B2026.0559.7752.9741.6
BaseBackbone=Qwen3-4B-Instruct-25072026.0559.753.3958.96