General Reasoning on ARC-E
98.7AccuracyLatentMAS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LatentMASBackbone=Qwen3-14B, MAS Setting=Hierarchical2025.11 | 98.7 | 619 | 1,884 | |
| LatentMASBackbone=Qwen3-8B, MAS Setting=Hierarchical2025.11 | 98.3 | 308 | 1,779 | |
| TextMASBackbone=Qwen3-14B, MAS Setting=Hierarchical2025.11 | 98.3 | 2,752 | 7,102 | |
| TextMASBackbone=Qwen3-8B, MAS Setting=Hierarchical2025.11 | 98.2 | 2,237 | 3,619 | |
| SingleBackbone=Qwen3-14B, MAS Setting=Hierarchical2025.11 | 97.2 | 608 | 551 | |
| TextMASBackbone=Qwen3-4B, MAS Setting=Hierarchical2025.11 | 97.1 | 2,054 | 2,239 | |
| LatentMASBackbone=Qwen3-4B, MAS Setting=Hierarchical2025.11 | 96.8 | 363 | 591 | |
| SingleBackbone=Qwen3-8B, MAS Setting=Hierarchical2025.11 | 95.6 | 656 | 404 | |
| SingleBackbone=Qwen3-4B, MAS Setting=Hierarchical2025.11 | 95.4 | 724 | 369 |