Web Browsing and Navigation on WebWalkerQA
71.7Average AccuracyOpenAI-o3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OpenAI-o32026.01 | 71.7 | — | — | — | |
| Claude-4-Sonnet2026.01 | 61.7 | — | — | — | |
| Reagent-UBackbone=Qwen3-8B2026.01 | 46.2 | 49.2 | 46.8 | 43.3 | |
| Reagent-RBackbone=Qwen3-8B2026.01 | 45.3 | 47.5 | 46 | 42.9 | |
| Reagent w/o Agent-RRMBackbone=Qwen3-8B2026.01 | 43.5 | 44.4 | 45 | 41.3 | |
| WebDancerBackbone=Qwen2.5-32B2026.01 | 38.4 | 44.3 | 46.7 | 29.2 | |
| WebDancerBackbone=Qwen2.5-7B2026.01 | 36 | 40.6 | 44.1 | 28.2 | |
| ARPOBackbone=Qwen3-14B2026.01 | 36 | 31.1 | 42.9 | 31 | |
| Reagent-CBackbone=Qwen3-8B, Inference Type=Direct Inference2026.01 | 35.5 | 35.6 | 38.1 | 32.4 | |
| Search-o1Backbone=QwQ-32B-Preview2026.01 | 34.1 | 43.1 | 35 | 27.1 | |
| ARPOBackbone=Qwen3-8B2026.01 | 30.5 | 26.7 | 33.3 | 29.6 | |
| Qwen3-8BBackbone=Qwen3-8B2026.01 | 29 | 31.1 | 28.6 | 28.2 | |
| Atom-SearcherBackbone=Qwen2.5-7B2026.01 | 27.9 | 31.7 | 23.7 | 37 | |
| WebThinkerBackbone=Qwen3-8B2026.01 | 13 | 6.7 | 13.1 | 16.9 | |
| Tree-GRPOBackbone=Qwen2.5-14B2026.01 | 12.8 | 11.1 | 15.5 | 10.8 | |
| DeepSeek-R1-671BBackbone=DeepSeek-R1-671B2026.01 | 10 | 5 | 11.8 | 11.3 | |
| o1-preview2026.01 | 9.9 | 11.9 | 10.4 | 7.9 | |
| QwQ-32BBackbone=QwQ-32B2026.01 | 3.8 | 7.5 | 2.1 | 4.2 |