Multi-Hop Question Answering on Bamboogle (OSR%)
37.1OSRStepSearch-base
Evaluation Results
| Method | Links | |
|---|---|---|
| StepSearch-baseBackbone=Qwen2.5-3B2026.04 | 37.1 | |
| StepSearch-instructBackbone=Qwen2.5-7B2026.04 | 32.3 | |
| Search-R1-baseBackbone=Qwen2.5-7B2026.04 | 30.6 | |
| Search-R1-instructBackbone=Qwen2.5-7B2026.04 | 29.8 | |
| StepSearch-baseBackbone=Qwen2.5-7B2026.04 | 29 | |
| Search-R1-instructBackbone=Qwen2.5-3B2026.04 | 16.9 | |
| StepSearch-instructBackbone=Qwen2.5-3B2026.04 | 15.3 | |
| HIPRAG-instructBackbone=Qwen2.5-7B2026.04 | 11.3 | |
| AutoSearchBackbone=Qwen2.5-7B2026.04 | 10.1 | |
| Search-R1-baseBackbone=Qwen2.5-3B2026.04 | 6.5 | |
| HIPRAG-baseBackbone=Qwen2.5-3B2026.04 | 3.2 | |
| HIPRAG-instructBackbone=Qwen2.5-3B2026.04 | 3.2 | |
| AutoSearchBackbone=Qwen2.5-3B2026.04 | 2.4 |