Reasoning on GPQA Diamond (Acc., Token, Speed)
53AccuracyLatentMAS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LatentMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-14B2025.11 | 53 | 3,606 | 1,458 | |
| LatentMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-14B2025.11 | 52 | 5,454 | 1,475 | |
| TextMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-14B2025.11 | 52 | 20,931 | 9,119 | |
| TextMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-14B2025.11 | 51.5 | 12,676 | 9,714 | |
| SingleMAS Setting=Sequential MAS Setting, Backbone=Qwen3-14B2025.11 | 48.5 | 5,547 | 1,043 | |
| SingleMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-14B2025.11 | 48.5 | 5,547 | 1,043 | |
| LatentMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-8B2025.11 | 46.9 | 3,395 | 798 | |
| LatentMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-8B2025.11 | 45.5 | 4,571 | 854 | |
| TextMASMAS Setting=Sequential MAS Setting, Backbone=Qwen3-8B2025.11 | 43.4 | 17,986 | 5,771 | |
| TextMASMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-8B2025.11 | 43 | 22,450 | 6,108 | |
| SingleMAS Setting=Sequential MAS Setting, Backbone=Qwen3-8B2025.11 | 39.9 | 6,435 | 813 | |
| SingleMAS Setting=Hierarchical MAS Setting, Backbone=Qwen3-8B2025.11 | 39.9 | 6,435 | 813 |