Reasoning on AIME25 (Accuracy, Token Usage, Speed)
0.667AccuracyTextMAS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TextMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-14B2025.11 | 0.667 | 50,003 | 6,019 | |
| LatentMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-14B2025.11 | 0.667 | 9,527 | 1,056 | |
| LatentMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-14B2025.11 | 0.633 | 11,402 | 1,473 | |
| TextMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-14B2025.11 | 0.6 | 44,618 | 5,184 | |
| SingleMAS Setting=Sequential MAS Setting, Backbone=Qwen3-14B2025.11 | 0.567 | 11,298 | 1,040 | |
| SingleMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-14B2025.11 | 0.567 | 11,298 | 1,040 | |
| TextMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-8B2025.11 | 0.533 | 45,088 | 3,150 | |
| LatentMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-8B2025.11 | 0.533 | 8,699 | 820 | |
| TextMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-8B2025.11 | 0.5 | 53,929 | 3,488 | |
| LatentMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-8B2025.11 | 0.5 | 13,230 | 616 | |
| SingleMAS Setting=Sequential MAS Setting, Backbone=Qwen3-8B2025.11 | 0.467 | 14,692 | 450 | |
| SingleMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-8B2025.11 | 0.467 | 14,692 | 450 |