List your company

Mathematical Reasoning on MATH (Pass@1, TC)

90Pass@1 AccuracyEAPO
61.71269.05676.483.744Jun 1, 2026

Evaluation results

MethodLinks
EAPOBackbone Model=Qwen2.5-7B-Instruct2026.06901.03
AEPOBackbone Model=Qwen2.5-7B-Instruct2026.0689.11.01
ARPOBackbone Model=Qwen2.5-7B-Instruct2026.0687.81.01
Reinforce++Backbone Model=Qwen2.5-7B-Instruct2026.0687.61.12
ToolStarBackbone Model=Qwen2.5-7B-Instruct2026.0686.41.28
GRPOBackbone Model=Qwen2.5-7B-Instruct2026.06860.97
BaseBackbone Model=Qwen2.5-7B-Instruct2026.0682—
Reinforce++Backbone Model=Qwen2.5-3B-Instruct2026.0681.81.12
EAPOBackbone Model=Qwen2.5-3B-Instruct2026.0681.61.01
ARPOBackbone Model=Qwen2.5-3B-Instruct2026.0680.81.05
GRPOBackbone Model=Qwen2.5-3B-Instruct2026.0680.51.07
ToolStarBackbone Model=Qwen2.5-3B-Instruct2026.0680.41.03
TIRBackbone Model=Qwen2.5-7B-Instruct2026.0678.21.54
BaseBackbone Model=Qwen2.5-3B-Instruct2026.0667.2—
TIRBackbone Model=Qwen2.5-3B-Instruct2026.0662.80.86