Multi-hop Question Answering on HotpotQA online Google Search API (test val)
51.3Exact Match (EM)ESearch-instruct
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ESearch-instructModel Scale=3B2025.10 | 51.3 | 61.2 | |
| ESearch-baseModel Scale=7B2025.10 | 51 | 63.2 | |
| Search-R1-baseModel Scale=7B2025.10 | 50.8 | 61 | |
| ESearch-instructModel Scale=7B2025.10 | 50.7 | 64.2 | |
| Research-instructModel Scale=7B2025.10 | 49.4 | 60.8 | |
| Search-R1-instructModel Scale=7B2025.10 | 46.4 | 57 | |
| StepSearch-instructModel Scale=7B2025.10 | 46.2 | 56 | |
| StepSearch-baseModel Scale=7B2025.10 | 44.6 | 55.2 | |
| ESearch-baseModel Scale=3B2025.10 | 43.5 | 58.6 | |
| ZeroSearch-baseModel Scale=7B2025.10 | 43.1 | 52.9 | |
| R-Search-instruct-PPOModel Scale=3B2025.10 | 39.8 | 49.5 | |
| StepSearch-instructModel Scale=3B2025.10 | 39.4 | 47 | |
| ZeroSearch-instructModel Scale=7B2025.10 | 39.4 | 48.3 | |
| SSRL-instructModel Scale=7B2025.10 | 38.8 | 46.5 | |
| Research-baseModel Scale=7B2025.10 | 38.6 | 48.6 | |
| R-Search-instruct-GRPOModel Scale=7B2025.10 | 37.6 | 46.8 | |
| R-Search-instruct-GRPOModel Scale=3B2025.10 | 37.4 | 46 | |
| R-Search-instruct-PPOModel Scale=7B2025.10 | 35.8 | 45.3 | |
| ZeroSearch-instructModel Scale=3B2025.10 | 35.7 | 45.3 | |
| Search-R1-instructModel Scale=3B2025.10 | 35 | 44.2 | |
| Search-R1-baseModel Scale=3B2025.10 | 34.8 | 43.1 | |
| SSRL-instructModel Scale=3B2025.10 | 34.6 | 42.4 | |
| StepSearch-baseModel Scale=3B2025.10 | 34.5 | 46.4 | |
| ZeroSearch-baseModel Scale=3B2025.10 | 32.4 | 41.4 |