Mathematics Reasoning on GSM8K (Acc, # Tokens)
96.36AccuracyCOPT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| COPTBackbone=Qwen3-8B, Reasoning Effort=High (Peak Accuracy)2026.05 | 96.36 | 1,813 | |
| COPTBackbone=Qwen3-8B, Reasoning Effort=Token-saving / Matched Accuracy2026.05 | 95.98 | 961 | |
| CoT (Greedy)Backbone=Qwen3-8B2026.05 | 95.83 | 2,240 | |
| CoTBackbone=Qwen3-8B2026.05 | 95.75 | 2,138 |