List your company

Mathematical Reasoning on GSM8K (Acc, Tok, ReL)

90.9AccuracyOurs
75.50879.50483.587.496Feb 1, 2026

Evaluation results

MethodLinks
OursBackbone=Qwen2.5-7B2026.0290.92,60387.2
Ours (Base)Backbone=Qwen2.5-7B2026.0289.42,64996.4
INFTYTHINKBackbone=Qwen2.5-7B2026.0287.93,237117
LightThinkerBackbone=Qwen2.5-7B2026.0285.62,77990
SepLLMBackbone=Qwen2.5-7B2026.0284.92,58292.3
H2OBackbone=Qwen2.5-7B2026.0283.52,73092
OursBackbone=Qwen2.5-1.5B2026.0282.12,11669.7
Ours (Base)Backbone=Qwen2.5-1.5B2026.0280.12,08676.2
INFTYTHINKBackbone=Qwen2.5-1.5B2026.0279.52,50389.3
LightThinkerBackbone=Qwen2.5-1.5B2026.0278.82,10969.6
SepLLMBackbone=Qwen2.5-1.5B2026.0277.32,23075.8
H2OBackbone=Qwen2.5-1.5B2026.0276.12,11671.7