Open-domain Question Answering on MuSiQue (ACC, SC)
26.2Accuracy (ACC)StepSearch
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| StepSearchBackbone=Qwen2.5-7B-Instruct2026.05 | 26.2 | 2.27 | |
| HiPRAGBackbone=Qwen2.5-7B-Instruct2026.05 | 23.8 | 2.55 | |
| RFTBackbone=Qwen2.5-7B-Instruct2026.05 | 23.1 | 2.5 | |
| SAASBackbone=Qwen2.5-7B-Instruct2026.05 | 22.6 | 1.3 | |
| SAASBackbone=Qwen2.5-3B-Instruct2026.05 | 20.9 | 1.62 | |
| RFTBackbone=Qwen2.5-3B-Instruct2026.05 | 17 | 2.76 | |
| HiPRAGBackbone=Qwen2.5-3B-Instruct2026.05 | 17 | 2.37 | |
| StepSearchBackbone=Qwen2.5-3B-Instruct2026.05 | 16.8 | 2.1 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.05 | 16.8 | 1.38 | |
| Search-R1Backbone=Qwen2.5-3B-Instruct2026.05 | 15.2 | 1.79 | |
| Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.05 | 8.8 | — | |
| Direct InferenceBackbone=Qwen2.5-3B-Instruct2026.05 | 6.1 | — |