Multi-Hop Question Answering on 2Wiki (Accuracy)
43.6Accuracy (2Wiki)ReasonRAG
Evaluation Results
| Method | Links | |
|---|---|---|
| ReasonRAGBackbone=Qwen2.5-7B2026.05 | 43.6 | |
| CriticSearchBackbone=Qwen2.5-7B2026.05 | 42.8 | |
| MultiSearchBackbone=Qwen2.5-7B, Model Variant=Base2026.05 | 41.6 | |
| MultiSearchBackbone=Qwen2.5-3B, Model Variant=Base2026.05 | 41.3 | |
| MultiSearchBackbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 41.2 | |
| CriticSearchBackbone=Qwen2.5-3B2026.05 | 40.9 | |
| AutoRefineBackbone=Qwen2.5-3B, Model Variant=Base2026.05 | 39.3 | |
| AdaSearchBackbone=Qwen2.5-3B2026.05 | 38.5 | |
| StepSearchBackbone=Qwen2.5-7B, Model Variant=Base2026.05 | 38.5 | |
| AutoRefineBackbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 38 | |
| AutoRefineBackbone=Qwen2.5-7B, Model Variant=Base2026.05 | 36.9 | |
| StepSearchBackbone=Qwen2.5-7B, Model Variant=Instruct2026.05 | 36.6 | |
| ZeroSearchBackbone=Qwen2.5-3B, Model Variant=Base2026.05 | 35.6 | |
| MultiSearchBackbone=Qwen2.5-7B, Model Variant=Instruct2026.05 | 35.1 | |
| Dr.ZeroBackbone=Qwen2.5-7B2026.05 | 34.7 | |
| Search-R1Backbone=Qwen2.5-7B, Model Variant=Instruct2026.05 | 34.6 | |
| ZeroSearchBackbone=Qwen2.5-7B, Model Variant=Base2026.05 | 34.6 | |
| StepSearchBackbone=Qwen2.5-3B, Model Variant=Base2026.05 | 33.9 | |
| AutoRefineBackbone=Qwen2.5-7B, Model Variant=Instruct2026.05 | 32.5 | |
| StepSearchBackbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 32 | |
| Search-R1Backbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 31 | |
| ZeroSearchBackbone=Qwen2.5-7B, Model Variant=Instruct2026.05 | 30 | |
| Search-R1Backbone=Qwen2.5-7B, Model Variant=Base2026.05 | 29.7 | |
| R1Backbone=Qwen2.5-7B, Model Variant=Instruct2026.05 | 29.2 | |
| Dr.ZeroBackbone=Qwen2.5-3B2026.05 | 29.1 | |
| R1Backbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 27.5 | |
| Search-R1Backbone=Qwen2.5-3B, Model Variant=Base2026.05 | 27.4 | |
| R1Backbone=Qwen2.5-7B, Model Variant=Base2026.05 | 27.3 | |
| ReSearchBackbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 27.2 | |
| ReSearchBackbone=Qwen2.5-3B, Model Variant=Base2026.05 | 27.2 | |
| R1Backbone=Qwen2.5-3B, Model Variant=Base2026.05 | 26.8 | |
| SFTBackbone=Qwen2.5-7B2026.05 | 25.9 | |
| Direct InferenceBackbone=Qwen2.5-7B2026.05 | 25 | |
| SFTBackbone=Qwen2.5-3B2026.05 | 24.8 | |
| Direct GenerationBackbone=Qwen2.5-3B2026.05 | 24.4 | |
| RAGBackbone=Qwen2.5-7B2026.05 | 23.5 | |
| RAGBackbone=Qwen2.5-3B2026.05 | 22.6 | |
| Search-o1Backbone=Qwen2.5-3B2026.05 | 21.8 | |
| ZeroSearchBackbone=Qwen2.5-3B, Model Variant=Instruct2026.05 | 21.4 | |
| Search-o1Backbone=Qwen2.5-7B2026.05 | 17.6 | |
| IRCoTBackbone=Qwen2.5-3B2026.05 | 17.1 | |
| IRCoTBackbone=Qwen2.5-7B2026.05 | 14.9 | |
| CoTBackbone=Qwen2.5-7B2026.05 | 11.1 | |
| CoTBackbone=Qwen2.5-3B2026.05 | 2.1 |