Deep-search QA on Xbench-DeepSearch (test)
75Pass@1Tongyi-DR-30B
Evaluation Results
| Method | Links | |
|---|---|---|
| Tongyi-DR-30BBackbone=30B, Type=Open-Source Agentic Foundation Model2026.02 | 75 | |
| DeepSeek-V3.1Type=Open-Source Agentic Foundation Model2026.02 | 71 | |
| MiroThinker-v1.0-30BBackbone=30B, Type=Open-Source Agentic Foundation Model2026.02 | 70.6 | |
| GLM-4.6Type=Open-Source Agentic Foundation Model2026.02 | 70 | |
| OpenAI-o3Type=Proprietary Agentic Foundation Model2026.02 | 67 | |
| Claude-4.5-SonnetType=Proprietary Agentic Foundation Model2026.02 | 66 | |
| QwQ-32B-RFT-Iter2 (SRR-Judge)Backbone=QwQ-32B, Training Strategy=Iterative RFT, Iteration=2, Supervision Type=SRR-Judge2026.02 | 61.3 | |
| Kimi-K2-0905Type=Open-Source Agentic Foundation Model2026.02 | 61 | |
| QwQ-32B-RFT-Iter1 (SRR-Judge)Backbone=QwQ-32B, Training Strategy=Iterative RFT, Iteration=1, Supervision Type=SRR-Judge2026.02 | 59.7 | |
| QwQ-32B-RFT-Iter1 (ORM)Backbone=QwQ-32B, Training Strategy=Iterative RFT, Iteration=1, Supervision Type=Outcome Supervision (ORM)2026.02 | 56.5 | |
| DeepSeek-R1 + SRR RefineBackbone=DeepSeek-R1, Refinement Method=SRR Refine2026.02 | 55.3 | |
| QwQ-32B-SFT + SRR RefineBackbone=QwQ-32B, Training Strategy=SFT, Refinement Method=SRR Refine2026.02 | 55 | |
| DeepSeek-R1Backbone=DeepSeek-R12026.02 | 53.7 | |
| WebSailor-32BBackbone=32B, Type=Open-Source Deep Search Solution2026.02 | 53.3 | |
| QwQ-32B + SRR RefineBackbone=QwQ-32B, Refinement Method=SRR Refine2026.02 | 53.3 | |
| QwQ-32B-SFTBackbone=QwQ-32B, Training Strategy=SFT2026.02 | 52.7 | |
| DeepDive-32BBackbone=32B, Type=Open-Source Deep Search Solution2026.02 | 51.8 | |
| Grok-DeepResearchType=Proprietary Agentic Foundation Model2026.02 | 50 | |
| Doubao-DeepThinkType=Proprietary Agentic Foundation Model2026.02 | 50 | |
| QwQ-32BBackbone=QwQ-32B2026.02 | 46 | |
| ASearcher-Web-32BBackbone=32B, Type=Open-Source Deep Search Solution2026.02 | 42.1 | |
| WebDancer-32BBackbone=32B, Type=Open-Source Deep Search Solution2026.02 | 39 | |
| Search-O1-32BBackbone=32B, Type=Open-Source Deep Search Solution2026.02 | 25 | |
| WebThinker-32BBackbone=32B, Type=Open-Source Deep Search Solution2026.02 | 24 |