Question Answering on Wiki 18 20M-scale file-per-document (50-query samples)
62NQ ScoreDR-DCI
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DR-DCIBackbone=GPT-5.4 Nano2026.06 | 62 | 82 | 64 | 68 | 58 | 44 | 63 | |
| R1-SearcherModel size=7B2026.06 | 58 | 50 | 54 | 46 | 40 | 24 | 45.33 | |
| Search-R1Model size=32B2026.06 | 56 | 46 | 52 | 44 | 50 | 32 | 46.67 | |
| Verl-Tool-Search-7B-DAPOModel size=7B, Optimization=DAPO2026.06 | 56 | 44 | 32 | 50 | 32 | 12 | 37.67 | |
| ASearcher-LocalModel size=14B2026.06 | 56 | 58 | 62 | 58 | 56 | 24 | 52.33 | |
| ZeroSearchModel size=7B2026.06 | 26 | 30 | 18 | 10 | 18 | 4 | 17.67 |