Search-Augmented Reasoning on GAIA Text
59L1 AccuracySAKE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SAKEBackbone=Qwen3-30B-A3B-Thinking2026.02 | 59 | 42.3 | 16.7 | 45.6 | |
| SAKEBackbone=Qwen3-4B-Thinking2026.02 | 56.4 | 55.8 | 8.33 | 50.5 | |
| Search-o1Backbone=Qwen3-30B-A3B-Thinking2026.02 | 56.4 | 32.7 | 16.7 | 39.8 | |
| SAKEBackbone=QwQ-32B2026.02 | 56.4 | 57.7 | 8.33 | 51.5 | |
| Search-o1Backbone=QwQ-32B2026.02 | 48.8 | 44.2 | 16.7 | 42.7 | |
| Search-o1Backbone=Qwen3-4B-Thinking2026.02 | 48.7 | 48.1 | 8.33 | 43.7 | |
| Direct Gen.Backbone=QwQ-32B2026.02 | 35.9 | 17.3 | 8.33 | 23.3 | |
| Direct Gen.Backbone=Qwen3-4B-Thinking2026.02 | 30.8 | 17.3 | 0 | 20.4 | |
| Direct Gen.Backbone=Qwen3-30B-A3B-Thinking2026.02 | 30.8 | 13.5 | 8.33 | 19.4 |