Web Navigation Question Answering on WebWalkerQA
58.8Success Rate (Easy)WebThinker-32B-RL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| WebThinker-32B-RLModel Category=Search-Agent-Trained Model2026.05 | 58.8 | 44.6 | 40.4 | 46.5 | |
| deep-research agentSearch setting=curated search setting (commercial retrieval services)2026.06 | 58.5 | 62.3 | 67.1 | 63.2 | |
| Hermes-SESearch setting=raw open-web setting (no pre-integrated search tools)2026.06 | 53.7 | 51.5 | 55.3 | 53 | |
| SkillHoneSearch setting=raw open-web setting (no pre-integrated search tools)2026.06 | 53.7 | 69.2 | 68.4 | 66.4 | |
| Existing-SkillsSearch setting=raw open-web setting (no pre-integrated search tools)2026.06 | 51.2 | 48.5 | 52.6 | 50.2 | |
| Qwen3-8B + ACTGUIDE-RLTraining=ACTGUIDE-RL2026.05 | 50 | 46.79 | 44.58 | 46.77 | |
| Qwen3-4B-Instruct + ACTGUIDE-RLTraining=ACTGUIDE-RL2026.05 | 43.75 | 41.78 | 35 | 39.85 | |
| Qwen3-8B + RLTraining=RL Baseline2026.05 | 43.75 | 44.64 | 39.16 | 42.5 | |
| APPOBackbone=Qwen3-14B, Method Strategy=APPO, Method Category=RL-based Method2026.06 | 43 | 48.6 | 37.6 | 43.4 | |
| Qwen3-8BTraining=Base2026.05 | 41.87 | 31.78 | 26.25 | 32.2 | |
| ARPOBackbone=Qwen3-14B, Method Strategy=ARPO, Method Category=RL-based Method2026.06 | 40 | 44.1 | 36.6 | 40.5 | |
| APPOBackbone=Qwen3-8B, Method Strategy=APPO, Method Category=RL-based Method2026.06 | 37.5 | 36.4 | 28.2 | 33.8 | |
| Skill-CreatorSearch setting=raw open-web setting (no pre-integrated search tools)2026.06 | 36.6 | 36.9 | 40.8 | 38.1 | |
| GRPOBackbone=Qwen3-14B, Method Strategy=GRPO, Method Category=RL-based Method2026.06 | 35.6 | 42.9 | 35.2 | 38.5 | |
| ARPOBackbone=Qwen3-8B, Method Strategy=ARPO, Method Category=RL-based Method2026.06 | 31.1 | 35.7 | 28.2 | 32 | |
| GRPOBackbone=Qwen3-8B, Method Strategy=GRPO, Method Category=RL-based Method2026.06 | 28.9 | 32.1 | 28.2 | 30 | |
| ARPO-8BModel Category=Search-Agent-Trained Model2026.05 | 26.7 | 33.3 | 29.6 | 30.5 | |
| Qwen2.5-7B-Instruct + ACTGUIDE-RLTraining=ACTGUIDE-RL2026.05 | 24.37 | 21.07 | 21.66 | 22.05 | |
| Qwen2.5-3B-Instruct + ACTGUIDE-RLTraining=ACTGUIDE-RL2026.05 | 18.75 | 16.07 | 22.08 | 18.82 | |
| Qwen2.5-7B-InstructTraining=Base2026.05 | 18.75 | 19.28 | 16.25 | 18.09 | |
| Vanilla RAGBackbone=Qwen3-14B, Method Strategy=Vanilla RAG, Method Category=Single-Enhanced Method2026.06 | 17.8 | 13.1 | 11.3 | 13.5 | |
| Qwen2.5-3B-Instruct + RLTraining=RL Baseline2026.05 | 15 | 15 | 15.83 | 15.29 | |
| Qwen2.5-7B-Instruct + RLTraining=RL Baseline2026.05 | 14.37 | 20.35 | 19.58 | 18.67 | |
| WebThinkerBackbone=Qwen3-14B, Method Strategy=WebThinker, Method Category=Single-Enhanced Method2026.06 | 13.3 | 23.8 | 18.3 | 19.5 | |
| Qwen3-4B-Instruct + RLTraining=RL Baseline2026.05 | 13.12 | 13.92 | 9.17 | 12.06 | |
| o1-previewBackbone=o1-preview, Method Category=Reference2026.06 | 11.9 | 10.4 | 7.9 | 9.9 | |
| Search-o1Backbone=Qwen3-14B, Method Strategy=Search-o1, Method Category=Single-Enhanced Method2026.06 | 11.1 | 21.4 | 16.9 | 17.5 | |
| ReActBackbone=Qwen3-14B, Method Strategy=ReAct, Method Category=Single-Enhanced Method2026.06 | 11.1 | 20.2 | 12.7 | 15.5 | |
| Vanilla RAGBackbone=Qwen3-8B, Method Strategy=Vanilla RAG, Method Category=Single-Enhanced Method2026.06 | 8.9 | 10.7 | 9.9 | 10 | |
| ReActBackbone=Qwen3-8B, Method Strategy=ReAct, Method Category=Single-Enhanced Method2026.06 | 8.9 | 16.7 | 18.3 | 15.5 | |
| Qwen3-4B-InstructTraining=Base2026.05 | 8.75 | 3.57 | 0.83 | 3.82 | |
| DeepSeek-R1-32BBackbone=DeepSeek-R1-32B, Method Category=Reference2026.06 | 7.5 | 1.4 | 4.2 | 3.8 | |
| QwQ-32BBackbone=QwQ-32B, Method Category=Reference2026.06 | 7.5 | 2.1 | 4.6 | 4.3 | |
| Qwen3-32B-thinkingBackbone=Qwen3-32B, Method Category=Reference2026.06 | 6.9 | 1.1 | 2.9 | 3.1 | |
| GPT-4oBackbone=GPT-4o, Method Category=Reference2026.06 | 6.7 | 6 | 4.2 | 5.5 | |
| Search-o1Backbone=Qwen3-8B, Method Strategy=Search-o1, Method Category=Single-Enhanced Method2026.06 | 6.7 | 15.5 | 9.7 | 11.5 | |
| WebThinkerBackbone=Qwen3-8B, Method Strategy=WebThinker, Method Category=Single-Enhanced Method2026.06 | 6.7 | 13.1 | 16.9 | 13 | |
| Qwen3-14BBackbone=Qwen3-14B, Method Strategy=Base, Method Category=RL-based Method2026.06 | 6.7 | 2.4 | 4.2 | 4 | |
| Qwen2.5-3B-InstructTraining=Base2026.05 | 5 | 7.14 | 4.58 | 5.73 | |
| DeepSeek-R1-671BBackbone=DeepSeek-R1-671B, Method Category=Reference2026.06 | 5 | 11.8 | 11.3 | 10 | |
| Qwen3-8BBackbone=Qwen3-8B, Method Strategy=Base, Method Category=RL-based Method2026.06 | 0 | 2.4 | 2.8 | 2 |