Question Answering on MuSiQue (F1, Acc, # Turn, AUC)
31.2F1 ScoreHybridDeepSearcher
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HybridDeepSearcherSearch Strategy=Hybrid Search (Ours), Backbone=Qwen3-8B2025.08 | 31.2 | 35.1 | 3.3 | 0.3 | |
| RAG-R1Search Strategy=Parallel Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 29.7 | 32.4 | 2.1 | 0.29 | |
| HybridDeepSearcherSearch Strategy=Hybrid Search (Ours), Backbone=Qwen2.5-7B-Instruct2025.08 | 28.1 | 32.6 | 2.8 | 0.26 | |
| Search-R1Search Strategy=Sequential Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 26.6 | 29.1 | 3.2 | 0.23 | |
| R1-SearcherSearch Strategy=Sequential Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 25.1 | 28.5 | 2.7 | 0.24 | |
| Search-o1Search Strategy=Sequential Search, Backbone=Qwen3-8B2025.08 | 23.4 | 31.8 | 3.7 | 0.26 | |
| DeepResearcherSearch Strategy=Parallel Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 21.7 | 23.4 | 3.4 | 0.19 | |
| Standard RAGSearch Strategy=Non-iterative, Backbone=Qwen3-8B2025.08 | 15.8 | 24.8 | — | — | |
| Naïve GenSearch Strategy=Non-iterative, Backbone=Qwen3-8B2025.08 | 12.8 | 16.4 | — | — |