Question Answering on xbench DeepSearch
56Accuracy (Pass@4)SearchGym-RL (Ours)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SearchGym-RL (Ours)Source=local, Backbone=Qwen-3-8B2026.01 | 56 | 5,200.55 | 5.13 | 1.81 | 0 | |
| SearchGym-RL (Ours-base)Source=local, Backbone=Qwen-2.5-7B-Base2026.01 | 49 | 1,114.48 | 3.71 | 0.9 | 0 | |
| ASearcher-base-localSource=local, Backbone=Qwen-2.5-7B-Base2026.01 | 43 | 1,022.38 | 7.03 | 0.03 | 0 | |
| ASearcher-base-webSource=web, Backbone=Qwen-2.5-7B-Base2026.01 | 32 | 692.28 | 5.92 | 0.07 | — |