Deep Search QA on MuSiQue
29.52AccuracyProCeedRL
Evaluation Results
| Method | Links | |
|---|---|---|
| ProCeedRLBackbone=Qwen3-8B2026.04 | 29.52 | |
| DeepResearcher2026.04 | 29.3 | |
| DAPO/Search-R1Backbone=Qwen3-8B2026.04 | 23.6 | |
| RFTBackbone=Qwen3-8B2026.04 | 22.49 | |
| Qwen3-8B-v3-SFTBackbone=Qwen3-8B2026.04 | 20.07 | |
| Rewinding MoreBackbone=Qwen3-8B, Type=ablation2026.04 | 19.11 | |
| GiGPO2026.04 | 18.9 | |
| ReAct PromptingBackbone=Qwen3-8B2026.04 | 18.64 |