Web Agent Search and Reasoning on xbench deepsearch
73.3AccuracyWebClipper (Hybrid)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| WebClipper (Hybrid)Category=Pruning Method, Base Model=Tongyi-DeepResearch, Variant=Hybrid2026.02 | 73.3 | 0.797 | 12.57 | 6,205 | |
| Tongyi-DeepResearchCategory=Open-sourced Agent, Evaluation Environment=Unified2026.02 | 71.3 | 0.779 | 14.26 | 6,918 | |
| WebClipper (Eff)Category=Pruning Method, Base Model=Tongyi-DeepResearch, Variant=Efficiency2026.02 | 71.3 | 0.792 | 10.81 | 5,931 | |
| Prompt ControlCategory=Pruning Method, Base Model=Tongyi-DeepResearch2026.02 | 67.6 | 0.763 | 12.5 | 6,321 | |
| OpenAI o3Category=Close-sourced System2026.02 | 67 | — | — | — | |
| Claude-4-SonnetCategory=Close-sourced System2026.02 | 64.6 | — | — | — | |
| Coarse PruneCategory=Pruning Method, Base Model=Tongyi-DeepResearch2026.02 | 60.3 | 0.725 | 8.85 | 4,774 | |
| Kimi-K2-Instruct-0905Category=Open-sourced Agent, Evaluation Environment=Unified2026.02 | 54 | 0.686 | 5.98 | 1,316 | |
| WebExplorerCategory=Open-sourced Agent2026.02 | 51.7 | 0.659 | 9.05 | 2,246 | |
| Qwen3-235B-A22B-Instruct-2507Category=Open-sourced Agent, Evaluation Environment=Unified2026.02 | 49 | 0.637 | 8.84 | 938 | |
| DeepSeek-R1-671BCategory=Open-sourced Agent, Evaluation Environment=Unified2026.02 | 42.7 | 0.59 | 4.38 | 1,941 |