Multi-hop Question Answering on DeepResearch-9K 29 (test)
0.45P-hatSPECHOP
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SPECHOPTarget Tool (T)=Web Search, Speculator (S)=GPT-4o2026.05 | 0.45 | 0.18 | 0.13 | 0.68 | 0.71 | |
| SPECHOPTarget Tool (T)=Web Search, Speculator (S)=GPT-4o mini2026.05 | 0.44 | 0.22 | 0.13 | 0.69 | 0.74 | |
| SPECHOPTarget Tool (T)=E5 Retrieval, Speculator (S)=Llama 3.1 8B2026.05 | 0.39 | 0.29 | 0.82 | 0.85 | 0.88 | |
| SPECHOPTarget Tool (T)=Web Search, Speculator (S)=Llama 3.1 8B2026.05 | 0.35 | 0.05 | 0.13 | 0.7 | 0.76 | |
| SPECHOPTarget Tool (T)=Web Search, Speculator (S)=Qwen 3 8B2026.05 | 0.31 | 0.05 | 0.13 | 0.74 | 0.78 | |
| SPECHOPTarget Tool (T)=E5 Retrieval, Speculator (S)=Qwen 3 8B2026.05 | 0.25 | 0.3 | 0.81 | 0.9 | 0.92 |