Web Navigation Task Completion on Mind2Web (Cross-website 177)
66.2Success RatePolySkill (+Update)
Evaluation Results
| Method | Links | |
|---|---|---|
| PolySkill (+Update)Backbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 66.2 | |
| ASI (+Update)Backbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 65.1 | |
| PolySkill (All Skills)Backbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 64.9 | |
| PolySkill (Same domain)Backbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 64.9 | |
| ASI (All skills)Backbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 64.8 | |
| ASI (Same domain)Backbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 64.8 | |
| BaselineBackbone=Claude-3.7-Sonnet, Training Tasks=1009 tasks2025.10 | 64.4 | |
| PolySkill (+Update)Backbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 61.3 | |
| PolySkill (Same domain)Backbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 58.9 | |
| ASI (+Update)Backbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 58.7 | |
| PolySkill (All Skills)Backbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 57.6 | |
| ASI (Same domain)Backbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 57 | |
| BaselineBackbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 56.2 | |
| ASI (All skills)Backbone=GPT-4.1, Training Tasks=1009 tasks2025.10 | 54.9 |