Math Problem Solving on MATH (Accuracy & Improvement)
76.4AccuracyPrimitives-based MAS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Primitives-based MASBackbone=Qwen3-14B2026.02 | 76.4 | 14.4 | |
| LatentMASBackbone=Qwen3-14B2026.02 | 72.2 | 10.2 | |
| VotingBackbone=Qwen3-14B2026.02 | 72 | 10 | |
| PlanningBackbone=Qwen3-14B2026.02 | 69.7 | 7.7 | |
| TextMASBackbone=Qwen3-14B2026.02 | 68.6 | 6.6 | |
| ReviewBackbone=Qwen3-14B2026.02 | 68.5 | 6.5 | |
| SingleBackbone=Qwen3-14B2026.02 | 62 | — | |
| Primitives-based MASBackbone=Qwen3-4B2026.02 | 60.5 | 6.4 | |
| LatentMASBackbone=Qwen3-4B2026.02 | 59.6 | 5.5 | |
| VotingBackbone=Qwen3-4B2026.02 | 56.6 | 2.5 | |
| PlanningBackbone=Qwen3-4B2026.02 | 55.6 | 1.5 | |
| TextMASBackbone=Qwen3-4B2026.02 | 55.4 | 1.3 | |
| ReviewBackbone=Qwen3-4B2026.02 | 54.9 | 0.8 | |
| SingleBackbone=Qwen3-4B2026.02 | 54.1 | — |