Mathematical Reasoning on GSM8K (Medium and Hard Subset)
27.6Recovery Rate (%)QA+
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| QA+Protocol=opus→opus→haiku2026.02 | 27.6 | 110 | |
| QAProtocol=haiku→opus→haiku2026.02 | 23.3 | 110 | |
| Bit-Limited Chain-of-Thought (BL-CoT)Protocol=haiku→haiku→haiku2026.02 | 19.1 | 110 |