Mathematical Reasoning on MATH-500 1.0 (test)
86.6AccuracyMUR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MURBackbone=Qwen3-30B-A3B, Search Strategy=LLM as a critic2025.07 | 86.6 | 1,594 | |
| MURBackbone=Qwen3-30B-A3B, Search Strategy=φ-Decoding2025.07 | 86 | 1,700 | |
| SMARTBackbone=Qwen3-30B-A3B, Search Strategy=Guided search2025.07 | 85.4 | 5,601 | |
| Per-Step ScaleBackbone=Qwen3-30B-A3B, Search Strategy=Guided search2025.07 | 85.2 | 7,345 | |
| Per-Step ScaleBackbone=Qwen3-30B-A3B, Search Strategy=LLM as a critic2025.07 | 84.8 | 2,985 | |
| MURBackbone=Qwen3-30B-A3B, Search Strategy=Guided search2025.07 | 84.6 | 1,564 | |
| Avg uncertaintyBackbone=Qwen3-30B-A3B, Search Strategy=LLM as a critic2025.07 | 84.4 | 2,502 | |
| Avg uncertaintyBackbone=Qwen3-30B-A3B, Search Strategy=Guided search2025.07 | 84.2 | 3,779 | |
| Vanilla CoTBackbone=Qwen3-30B-A3B2025.07 | 83.4 | 606 | |
| SMARTBackbone=Qwen3-30B-A3B, Search Strategy=LLM as a critic2025.07 | 83.4 | 1,057 | |
| SMARTBackbone=Qwen3-30B-A3B, Search Strategy=φ-Decoding2025.07 | 82.8 | 3,052 | |
| Per-Step ScaleBackbone=Qwen3-30B-A3B, Search Strategy=φ-Decoding2025.07 | 82.6 | 3,609 | |
| Avg uncertaintyBackbone=Qwen3-30B-A3B, Search Strategy=φ-Decoding2025.07 | 81.2 | 1,990 | |
| Per-Step ScaleBackbone=GLM-4-9B, Search Strategy=Guided search2025.07 | 56.6 | 3,494 | |
| Avg uncertaintyBackbone=GLM-4-9B, Search Strategy=LLM as a critic2025.07 | 56 | 1,008 | |
| SMARTBackbone=GLM-4-9B, Search Strategy=Guided search2025.07 | 53.4 | 2,304 | |
| Per-Step ScaleBackbone=GLM-4-9B, Search Strategy=LLM as a critic2025.07 | 53.2 | 1,037 | |
| MURBackbone=GLM-4-9B, Search Strategy=LLM as a critic2025.07 | 53.2 | 746 | |
| MURBackbone=GLM-4-9B, Search Strategy=Guided search2025.07 | 49.6 | 901 | |
| Avg uncertaintyBackbone=GLM-4-9B, Search Strategy=Guided search2025.07 | 48.6 | 1,196 | |
| SMARTBackbone=GLM-4-9B, Search Strategy=LLM as a critic2025.07 | 48.6 | 579 | |
| MURBackbone=GLM-4-9B, Search Strategy=φ-Decoding2025.07 | 48 | 1,531 | |
| Avg uncertaintyBackbone=GLM-4-9B, Search Strategy=φ-Decoding2025.07 | 47.8 | 1,316 | |
| SMARTBackbone=GLM-4-9B, Search Strategy=φ-Decoding2025.07 | 47 | 2,386 | |
| Vanilla CoTBackbone=GLM-4-9B2025.07 | 45.2 | 473 | |
| Per-Step ScaleBackbone=GLM-4-9B, Search Strategy=φ-Decoding2025.07 | 44.6 | 3,817 |