List your company

Mathematics Reasoning on GSM8K (Acc, # Tokens)

96.36AccuracyCOPT
95.725695.890396.05596.2197May 19, 2026

Evaluation results

MethodLinks
COPTBackbone=Qwen3-8B, Reasoning Effort=High (Peak Accuracy)2026.0596.361,813
COPTBackbone=Qwen3-8B, Reasoning Effort=Token-saving / Matched Accuracy2026.0595.98961
CoT (Greedy)Backbone=Qwen3-8B2026.0595.832,240
CoTBackbone=Qwen3-8B2026.0595.752,138