List your company

Mathematical Reasoning on MathQA (Accuracy, ECE, Brier)

80.2AccuracySFT
57.94463.72269.575.278Jun 12, 2026

Evaluation results

MethodLinks
SFTBackbone Model=Ministral 3 8B2026.0680.212.429.51
CoRABackbone Model=Ministral 3 8B2026.0680.26.5822.66
GRPOBackbone Model=Ministral 3 8B2026.0679.87.8825.86
BaseBackbone Model=Ministral 3 8B2026.0679.48.3526.43
GRPOBackbone Model=Qwen2.5 7B2026.0677.213.9435.5
CoRABackbone Model=Qwen2.5 7B2026.0676.411.0932.83
SFTBackbone Model=Qwen2.5 7B2026.0672.817.5541.69
BaseBackbone Model=Qwen2.5 7B2026.067211.5537.69
CoRABackbone Model=Gemma 2 9B Instruct2026.0663.425.658.27
GRPOBackbone Model=Gemma 2 9B Instruct2026.0661.829.964.7
SFTBackbone Model=Gemma 2 9B Instruct2026.0661.234.4370.67
BaseBackbone Model=Gemma 2 9B Instruct2026.0658.832.2168.08