Web Agent Navigation on Mind2Web Cross-Task 1.0
49.2Success RatePolySkill (+Update)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PolySkill (+Update)Backbone=GLM-4.5, Training tasks=1009 tasks2025.10 | 49.2 | — | — | |
| PolySkill (+Update)Backbone=Qwen3-Coder-480B-A35B, Training tasks=1009 tasks2025.10 | 47.5 | — | — | |
| ASI (+Update)Backbone=GLM-4.5, Training tasks=1009 tasks2025.10 | 47 | — | — | |
| PolySkillBackbone=GLM-4.5, Training tasks=1009 tasks2025.10 | 46.1 | — | — | |
| ASIBackbone=GLM-4.5, Training tasks=1009 tasks2025.10 | 45.2 | — | — | |
| ASI (+Update)Backbone=Qwen3-Coder-480B-A35B, Training tasks=1009 tasks2025.10 | 44.9 | — | — | |
| BaselineBackbone=GLM-4.5, Training tasks=1009 tasks2025.10 | 44.8 | — | — | |
| PolySkillBackbone=Qwen3-Coder-480B-A35B, Training tasks=1009 tasks2025.10 | 42.3 | — | — | |
| ASIBackbone=Qwen3-Coder-480B-A35B, Training tasks=1009 tasks2025.10 | 41.5 | — | — | |
| BaselineBackbone=Qwen3-Coder-480B-A35B, Training tasks=1009 tasks2025.10 | 40.8 | — | — | |
| DuSARModel=Llama3.1-70B2025.12 | 5.28 | 54.9 | 45.1 | |
| DuSARModel=Gemma3-12B2025.12 | 5.21 | 53.2 | 45.2 | |
| TRADModel=Llama3.1-70B2025.12 | 2.38 | 44.2 | 37.6 | |
| DuSARModel=Llama3.1-8B2025.12 | 1.98 | 48.2 | 34.3 | |
| TRADModel=Gemma3-12B2025.12 | 1.98 | 34.8 | 31.1 | |
| SynapseModel=Gemma3-12B2025.12 | 1.19 | 31 | 26.1 | |
| SynapseModel=Llama3.1-70B2025.12 | 1.19 | 35 | 22 | |
| ReActModel=Gemma3-12B2025.12 | 0.4 | 23.4 | 17.3 | |
| ReActModel=Llama2-7B2025.12 | 0 | 7.2 | 6 | |
| SynapseModel=Llama2-7B2025.12 | 0 | 12.3 | 9.6 | |
| TRADModel=Llama2-7B2025.12 | 0 | 11.4 | 9.1 | |
| DuSARModel=Llama2-7B2025.12 | 0 | 15.1 | 12.6 | |
| ReActModel=Llama3.1-8B2025.12 | 0 | 7.1 | 3.5 | |
| SynapseModel=Llama3.1-8B2025.12 | 0 | 22.7 | 9.8 | |
| TRADModel=Llama3.1-8B2025.12 | 0 | 17.4 | 13.6 | |
| ReActModel=Llama3.1-70B2025.12 | 0 | 15.2 | 12 |