Reasoning on ARC Challenge (Accuracy, Token, Speed)
95.9AccuracyTextMAS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TextMASBackbone=Qwen3-14B, Collaboration Method=Text-based MAS, MAS Setting=Sequential2025.11 | 95.9 | 2,985 | 5,125 | |
| LatentMASBackbone=Qwen3-14B, Collaboration Method=Latent-based MAS, MAS Setting=Sequential2025.11 | 95.6 | 426 | 1,136 | |
| TextMASBackbone=Qwen3-8B, Collaboration Method=Text-based MAS, MAS Setting=Sequential2025.11 | 94.6 | 2,252 | 2,059 | |
| LatentMASBackbone=Qwen3-8B, Collaboration Method=Latent-based MAS, MAS Setting=Sequential2025.11 | 94.4 | 529 | 703 | |
| SingleBackbone=Qwen3-14B, Collaboration Method=Single-model, MAS Setting=Sequential2025.11 | 92.6 | 773 | 338 | |
| LatentMASBackbone=Qwen3-4B, Collaboration Method=Latent-based MAS, MAS Setting=Sequential2025.11 | 92.3 | 718 | 260 | |
| SingleBackbone=Qwen3-8B, Collaboration Method=Single-model, MAS Setting=Sequential2025.11 | 91 | 846 | 266 | |
| TextMASBackbone=Qwen3-4B, Collaboration Method=Text-based MAS, MAS Setting=Sequential2025.11 | 90 | 2,678 | 1,579 | |
| SingleBackbone=Qwen3-4B, Collaboration Method=Single-model, MAS Setting=Sequential2025.11 | 89.2 | 913 | 97 |