Multi-Hop Question Answering on Bamboogle (test) (Pass@3)
59.2Pass@3PivoARL
Evaluation Results
| Method | Links | |
|---|---|---|
| PivoARLBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 59.2 | |
| MetaRLBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 53.6 | |
| StepSearchBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 51.2 | |
| GiGPOBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 48 | |
| GRPOBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 47.2 | |
| Search-R1Base Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 46.4 | |
| ZeroSearchBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 46.4 | |
| GSPOBase Model=Qwen3-4B, Search-augmented=true, Domain=Out-of-domain2026.07 | 40.8 |