List your company

Mathematics Reasoning on Math500 (Accuracy, Token Count)

97.6Accuracy (%)COPT
95.93696.36896.897.232May 19, 2026

Evaluation results

MethodLinks
COPTBackbone=Qwen3-8B, Reasoning Effort=High (Peak Accuracy)2026.0597.64,851
CoT (Greedy)Backbone=Qwen3-8B2026.0596.45,311
COPTBackbone=Qwen3-8B, Reasoning Effort=Token-saving / Matched Accuracy2026.0596.23,609
CoTBackbone=Qwen3-8B2026.05964,985