Mathematical Reasoning on (Math500, Gaokao En, Olympiad, GSM8K, AMC23, AIME25, AIME24)
92.8Math500 ScoreDisagreement-Guided Strategy Routing
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Disagreement-Guided Strategy RoutingReasoning Strategy=ours, Backbone=Qwen3-8B, Context Setting=8K-context2026.04 | 92.8 | 82.9 | 61.9 | 95.8 | 87.5 | 43.3 | 66.7 | 75.8 | |
| Disagreement-Guided Strategy RoutingReasoning Strategy=ours, Backbone=Qwen3-4B, Context Setting=8K-context2026.04 | 90.6 | 81.8 | 57.9 | 94.5 | 87.5 | 43.3 | 56.7 | 73.2 | |
| Majority VotingReasoning Strategy=majority voting, Backbone=Qwen3-4B, Context Setting=8K-context2026.04 | 88.6 | 79.5 | 53.2 | 95.5 | 82.5 | 26.7 | 50 | 68 | |
| DV (Dynamic Voting)Reasoning Strategy=dynamic voting, Backbone=Qwen3-4B, Context Setting=8K-context2026.04 | 88.6 | 78.4 | 51.3 | 95.1 | 80 | 30 | 46.7 | 67.2 | |
| Disagreement-Guided Strategy RoutingReasoning Strategy=ours, Backbone=DeepSeek-R1-Distill-Llama-8B, Context Setting=8K-context2026.04 | 88.4 | 81.3 | 54.7 | 91.2 | 87.5 | 33 | 50 | 69.4 | |
| SCoPReasoning Strategy=SCoP, Backbone=Qwen3-8B, Context Setting=8K-context2026.04 | 88.2 | 79.5 | 52.6 | 95.4 | 82.5 | 33.3 | 40 | 67.4 | |
| SCoPReasoning Strategy=SCoP, Backbone=DeepSeek-R1-Distill-Llama-8B, Context Setting=8K-context2026.04 | 88.2 | 78.2 | 51.9 | 89 | 85 | 33.3 | 40 | 66.5 | |
| SCoPReasoning Strategy=SCoP, Backbone=Qwen3-4B, Context Setting=8K-context2026.04 | 87.8 | 80.3 | 55.6 | 93.5 | 82.5 | 26.7 | 40 | 66.6 | |
| Majority VotingReasoning Strategy=majority voting, Backbone=Qwen3-8B, Context Setting=8K-context2026.04 | 87.6 | 79.7 | 50.7 | 95.8 | 85 | 33.3 | 50 | 68.9 | |
| DV (Dynamic Voting)Reasoning Strategy=dynamic voting, Backbone=Qwen3-8B, Context Setting=8K-context2026.04 | 87.6 | 79.7 | 49.5 | 95.7 | 85 | 30 | 43.3 | 67.3 | |
| BoNReasoning Strategy=BoN, Backbone=Qwen3-4B, Context Setting=8K-context2026.04 | 87.4 | 76.4 | 50.1 | 94.1 | 72.5 | 26.7 | 50 | 65.3 | |
| BoNReasoning Strategy=BoN, Backbone=Qwen3-8B, Context Setting=8K-context2026.04 | 85.8 | 75.6 | 49 | 95.1 | 65 | 33.3 | 40 | 63.4 | |
| DV (Dynamic Voting)Reasoning Strategy=dynamic voting, Backbone=DeepSeek-R1-Distill-Llama-8B, Context Setting=8K-context2026.04 | 85.8 | 76.9 | 51 | 87.5 | 77.5 | 33.3 | 46.7 | 65.5 | |
| Qwen3-4BReasoning Strategy=baseline, Backbone=Qwen3-4B, Context Setting=8K-context2026.04 | 85.4 | 75.6 | 49 | 94.2 | 65 | 23.3 | 30 | 60.4 | |
| Majority VotingReasoning Strategy=majority voting, Backbone=DeepSeek-R1-Distill-Llama-8B, Context Setting=8K-context2026.04 | 84.6 | 79.8 | 51.6 | 87.5 | 80 | 33.3 | 46.7 | 66.2 | |
| Qwen3-8BReasoning Strategy=baseline, Backbone=Qwen3-8B, Context Setting=8K-context2026.04 | 84.4 | 77.7 | 46.4 | 95.7 | 65 | 23.3 | 36.7 | 61.3 | |
| DS-Llama-8BReasoning Strategy=baseline, Backbone=DeepSeek-R1-Distill-Llama-8B, Context Setting=8K-context2026.04 | 80.2 | 74 | 46.4 | 85 | 67.5 | 30 | 40 | 60.4 | |
| BoNReasoning Strategy=BoN, Backbone=DeepSeek-R1-Distill-Llama-8B, Context Setting=8K-context2026.04 | 79 | 78.8 | 55.3 | 88.8 | 75 | 30 | 43.3 | 64.3 |