List your company

Reasoning on GPQA-Diamond, AIME2024, MATH500, and HumanEval

85.77Average ScoreVanilla
72.125275.667679.2182.7524Oct 24, 2025

Evaluation results

MethodLinks
VanillaBase Model=Qwen3-32B2025.1085.7765.6681.6797.698.17
Safety RecompositionBase Model=Qwen3-32B2025.1084.9162.3882.0897.697.56
VanillaBase Model=Qwen3-14B2025.1083.663.1477.9297.695.73
Safety BacktrackBase Model=Qwen3-32B2025.1083.5761.6277.0897.498.17
Safety BacktrackBase Model=Qwen3-14B2025.1083.3462.1277.929796.34
Safety RecompositionBase Model=Qwen3-14B2025.1083.2160.3678.7597.496.34
VanillaBase Model=Qwen3-8B2025.1081.2857.3377.597.692.68
Safety BacktrackBase Model=Qwen3-8B2025.1080.7854.377.597.493.9
Safety RecompositionBase Model=Qwen3-8B2025.1079.8956.8276.2592.693.9
STAR-1Base Model=Qwen3-14B2025.1079.6356.3276.2588.497.56
STAR-1Base Model=Qwen3-8B2025.1079.5757.3371.2596.493.29
SafeChainBase Model=Qwen3-14B2025.1077.9957.5871.2587.495.73
SafePathBase Model=Qwen3-8B2025.1077.5955.5667.9294.892.07
SafeChainBase Model=Qwen3-32B2025.1077.1954.371.786.496.34
STAR-1Base Model=Qwen3-32B2025.1076.9554.5572.9285.295.12
SafeChainBase Model=Qwen3-8B2025.1075.9952.2866.2595.290.24
SafeKeyBase Model=Qwen3-14B2025.1075.174976.788.486.58
SafeKeyBase Model=Qwen3-32B2025.107554.371.786.887.2
SafePathBase Model=Qwen3-14B2025.1073.8557.7870.4278.888.41
SafeKeyBase Model=Qwen3-8B2025.1073.4841.970.589091.46
SafePathBase Model=Qwen3-32B2025.1072.6562.3870.2560.497.56