Mathematical Reasoning on MATH Num. Thy All Non-Easy
43.7Recovery RateQA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| QAProtocol=haiku→opus→haiku2026.02 | 43.7 | 61 | 580 | 0.001 | |
| QA+Protocol=opus→opus→haiku2026.02 | 40.4 | 61 | — | — | |
| Bit-Limited Chain-of-Thought (BL-CoT)Protocol=haiku→haiku→haiku2026.02 | 38.3 | 61 | — | — |