Mathematics Reasoning on Math500 (Accuracy, Token Count)
97.6Accuracy (%)COPT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| COPTBackbone=Qwen3-8B, Reasoning Effort=High (Peak Accuracy)2026.05 | 97.6 | 4,851 | |
| CoT (Greedy)Backbone=Qwen3-8B2026.05 | 96.4 | 5,311 | |
| COPTBackbone=Qwen3-8B, Reasoning Effort=Token-saving / Matched Accuracy2026.05 | 96.2 | 3,609 | |
| CoTBackbone=Qwen3-8B2026.05 | 96 | 4,985 |