Multi-Hop Question Answering on Bamboogle (out-of-domain)
73.8AccuracySKILLRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SKILLRLBackbone=Qwen2.5-7B-Instruct2026.02 | 73.8 | — | |
| WebFilterEnvironment=Web Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 73.1 | 74.3 | |
| DeepResearcherEnvironment=Web Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 71 | 72.8 | |
| R1-SearcherEnvironment=Web Search, Backbone=Qwen2.5-7B-Instruct2025.08 | 64.8 | 65.6 | |
| Search-r1-baseEnvironment=Local RAG2025.08 | 56.6 | 57.6 | |
| EvolveRBackbone=Qwen2.5-7B-Instruct2026.02 | 54.4 | — | |
| GPT-4oEnvironment=Direct Reasoning2025.08 | 52.6 | 43.2 | |
| Search-o1Environment=Local RAG2025.08 | 46.6 | 53.6 | |
| Search-r1-instructEnvironment=Local RAG2025.08 | 45 | 47.2 | |
| StepSearchBackbone=Qwen2.5-7B-Instruct2026.02 | 40 | — | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.02 | 36.8 | — | |
| Gemini-2.0-FlashEnvironment=Direct Reasoning2025.08 | 36.5 | 28 | |
| Search-o1Backbone=Qwen2.5-7B-Instruct2026.02 | 30.4 | — | |
| ZeroSearchBackbone=Qwen2.5-7B-Instruct2026.02 | 27.8 | — | |
| Qwen3-32BEnvironment=Direct Reasoning2025.08 | 24.7 | 18.4 | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.02 | 24 | — | |
| R1-InstructBackbone=Qwen2.5-7B-Instruct2026.02 | 19.2 | — | |
| RAGBackbone=Qwen2.5-7B-Instruct2026.02 | 16.8 | — | |
| Qwen2.5Backbone=Qwen2.5-7B-Instruct2026.02 | 14.4 | — |