Deep search QA on WebWalkerQA
23.01AccuracyProCeedRL
Evaluation Results
| Method | Links | |
|---|---|---|
| ProCeedRLBackbone=Qwen3-8B2026.04 | 23.01 | |
| Rewinding MoreBackbone=Qwen3-8B, Type=ablation2026.04 | 20.16 | |
| Qwen3-8B-v3-SFTBackbone=Qwen3-8B2026.04 | 19.85 | |
| RFTBackbone=Qwen3-8B2026.04 | 19.56 | |
| DAPO/Search-R1Backbone=Qwen3-8B2026.04 | 19.51 | |
| ReAct PromptingBackbone=Qwen3-8B2026.04 | 18.62 |