Web Navigation Task Success on MIND2WEB ONLINE (test)
67Task Success Rate (Overall)Orchard-GUI-4B (SFT + RL)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Orchard-GUI-4B (SFT + RL)# Steps=30, # Tasks=2.6k2026.05 | 67 | — | — | — | |
| NavigatorMain Model=n1-preview-11-2025, Provider=Yutori, Open Source=false2026.02 | 64.7 | 84 | 62.2 | 48.7 | |
| Qwen3-VL-235B-A22B-Thinking# Steps=302026.05 | 63.7 | — | — | — | |
| OperatorMain Model=OpenAI Computer-Using Agent, Provider=OpenAI, Open Source=false2026.02 | 58.3 | 73.5 | 59.4 | 39.2 | |
| OpenAI computer-use-preview# Steps=1002026.05 | 58.3 | — | — | — | |
| GPT-5 (SoM)# Steps=1002026.05 | 57.7 | — | — | — | |
| Google Computer UseMain Model=Gemini 2.5 Computer Use, Provider=Google DeepMind, Open Source=false2026.02 | 57.3 | 77.1 | 55.2 | 45.9 | |
| Gemini computer-use-preview# Steps=1002026.05 | 57.3 | — | — | — | |
| o3 (SoM)# Steps=1002026.05 | 55.4 | — | — | — | |
| AVENIR-WEBMain Model=Gemini 3 Pro, Provider=Ours, Open Source=true2026.02 | 53.7 | 74.1 | 54.6 | 30.3 | |
| ACT-1-20250814Main Model=o3-2025-04-16 and Claude-sonnet-4-20250514, Provider=Enhans, Open Source=false2026.02 | 52.7 | 71.1 | 52.4 | 32.4 | |
| WebNavigatorModel=Gemini-2.5-Pro, Act #=62026.03 | 52.7 | — | — | — | |
| Claude Computer Use 3.7Main Model=Claude-3.7-sonnet-20250219, Provider=Anthropic, Open Source=false2026.02 | 47.3 | 75.9 | 41.3 | 27 | |
| Orchard-GUI-4B-SFT# Steps=30, # Tasks=0.4k2026.05 | 47 | — | — | — | |
| Gemini-3-flash (Axtree)# Steps=1002026.05 | 44.8 | — | — | — | |
| GPT-5 (Axtree)# Steps=302026.05 | 41.9 | — | — | — | |
| WebNavigatorModel=GPT-4o, Act #=62026.03 | 41.3 | — | — | — | |
| WebNavigatorModel=Qwen3-VL-32B-Instruct, Act #=62026.03 | 39.7 | — | — | — | |
| ACT-1-20250703Main Model=o3-2025-04-16 and Claude-sonnet-4-20250514, Provider=Enhans, Open Source=false2026.02 | 39.5 | 53.7 | 39.2 | 24.3 | |
| WebNavigatorModel=Claude-Sonnet-4, Act #=62026.03 | 38.7 | — | — | — | |
| MolmoWeb-8B# Steps=100, # Tasks=>278.5K2026.05 | 35.3 | — | — | — | |
| WebDreamerModel=GPT-4o, Act #=122026.03 | 35 | — | — | — | |
| Gemini-3-flash (Axtree)# Steps=302026.05 | 34.8 | — | — | — | |
| GPT-4o (SoM)# Steps=1002026.05 | 34.6 | — | — | — | |
| Fara-7B# Steps=100, # Tasks=>123.2k2026.05 | 34.1 | — | — | — | |
| GLM-4.1V-9B-Thinking# Steps=1002026.05 | 33.9 | — | — | — | |
| Qwen3-VL-4B-Thinking# Steps=302026.05 | 32 | — | — | — | |
| UI-TARS-1.5-7B# Steps=1002026.05 | 31.3 | — | — | — | |
| MolmoWeb-4B# Steps=100, # Tasks=>278.5k2026.05 | 31.3 | — | — | — | |
| SeeActMain Model=gpt-4o-2024-08-06, Provider=OSU, Open Source=true2026.02 | 30 | 51.8 | 28 | 9.5 | |
| SeeActBackbone=GPT-42026.04 | 28 | — | — | — | |
| Agent-EMain Model=gpt-4o-2024-08-06, Provider=Emergence AI, Open Source=true2026.02 | 27 | 51.8 | 23.1 | 6.8 | |
| Agent-eModel=GPT-4o, Act #=82026.03 | 27 | — | — | — | |
| PageGuideBackbone=google/gemini-3-flash-preview2026.04 | 26.57 | — | — | — | |
| Browser UseMain Model=gpt-4o-2024-08-06, Provider=Browser Use, Open Source=true2026.02 | 26 | 44.6 | 23.1 | 10.8 | |
| AVENIR-WEBMain Model=Qwen-3-VL-8B, Provider=Ours, Open Source=true2026.02 | 25.7 | 42 | 23.8 | 11.8 | |
| PageGuideBackbone=google/gemini-2.5-flash-preview2026.04 | 24.48 | — | — | — | |
| Claude 3.5Main Model=Claude-3-5-sonnet-20241022, Provider=Anthropic, Open Source=false2026.02 | 24 | 51.8 | 16.1 | 8.1 | |
| PageGuideBackbone=google/gemini-3-flash-preview2026.04 | 18.18 | — | — | — | |
| PageGuideBackbone=google/gemini-2.5-flash-preview2026.04 | 14.29 | — | — | — | |
| SeeActBackbone=GPT-42026.04 | 9.5 | — | — | — | |
| InfiniteWebTraining Tasks=2002026.01 | — | 54.3 | 18.2 | 7.9 | |
| InfiniteWebTraining Tasks=4002026.01 | — | 48.2 | 27.3 | 5.3 | |
| InfiniteWebTraining Tasks=6002026.01 | — | 56.8 | 23.1 | 9.2 | |
| Original BaselineTraining Tasks=Orig.2026.01 | — | 46.9 | 18.9 | 5.3 |