Web Agent Tasks on Mind2Web Cross-Domain
55.7Ele.AccMagma-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Magma-8BBackbone=LLaMA3 [92], DoM Tree=true2025.02 | 55.7 | 80.6 | 47.3 | |
| OmniParser2024.11 | 45.5 | 85.7 | 42 | |
| GPT-4V-OmniParserBackbone=GPT-4V [99], DoM Tree=true, Image=true2025.02 | 45.5 | 85.7 | 42 | |
| SKILL.nbZero-shot=true2026.06 | 44.9 | 46 | 39.7 | |
| SeeAct (GPT-4V)Backbone=GPT-4V [99], Image=true2025.02 | 42.4 | 69.3 | 36.8 | |
| MindActw/o HTML=false2024.01 | 42.1 | 66.5 | 39.6 | |
| MindActrequired HTML text inputs=true2024.11 | 42.1 | 66.5 | 39.6 | |
| AWMonlineZero-shot=true2026.06 | 40.9 | 36.3 | 35.5 | |
| ReasoningBankZero-shot=true2026.06 | 40.6 | 41.3 | 36.6 | |
| Qwen2-UIXBackbone=Qwen2 [124]2025.02 | 40.4 | — | 34.9 | |
| ShowUIFT?=true2024.11 | 39.4 | 86.8 | 35.2 | |
| ShowUI †FT?=true, Action history only=true2024.11 | 38.9 | 85.3 | 34.1 | |
| GPT-4w/o HTML=false2024.01 | 37.1 | 46.5 | 26.4 | |
| GPT-4close-source=true2024.11 | 37.1 | 46.5 | 26.4 | |
| GPT-4-MindActBackbone=GPT-4 [98], Image=true2025.02 | 37.1 | 46.5 | 26.4 | |
| Qwen2-VL-2BFT?=true2024.11 | 36.3 | 81.8 | 30.7 | |
| CodeActZero-shot=true2026.06 | 35.8 | 37.9 | 31.9 | |
| CogAgentBackbone=CogVLM [118]2025.02 | 33.1 | — | 28.5 | |
| ShowUI-ZSFT?=false2024.11 | 24.4 | 83.9 | 21.4 | |
| DAVELanguage Model Backbone=Llama-3.2-3B-Instruct2025.12 | 23.9 | — | 19.1 | |
| SeeClickw/o HTML=true2024.01 | 23.2 | 84.8 | 20.8 | |
| SeeClickFT?=true2024.11 | 23.2 | 84.8 | 20.8 | |
| SeeClickBackbone=Qwen-VL [3]2025.02 | 23.2 | 84.8 | 20.8 | |
| GPT-3.5-Turbow/o HTML=false2024.01 | 21.6 | 52.8 | 18.6 | |
| SeeAct (Gemini-Pro)Backbone=Gemini-Pro [36], Image=true2025.02 | 20.7 | 64.3 | 18 | |
| CogAgentFT?=true2024.11 | 20.6 | 42 | 15.5 | |
| DolphinLanguage Model Backbone=Llama-3.2-3B-Instruct2025.12 | 19.3 | — | 14.5 | |
| MiniCPM-V-GUIBackbone=MiniCPM-V [128], DoM Tree=true2025.02 | 17.9 | 74.5 | 17.6 | |
| MAELanguage Model Backbone=Llama-3.2-3B-Instruct2025.12 | 15.5 | — | 11.6 | |
| MindAct (gen)w/o HTML=false2024.01 | 14.2 | 44.7 | 11.9 | |
| Fuyu-8B-GUIBackbone=Fuyu-8B [4], DoM Tree=true2025.02 | 14.2 | 83.1 | 11.7 | |
| Qwen-VLw/o HTML=true2024.01 | 14.1 | 84.3 | 12 | |
| Qwen-VLFT?=true2024.11 | 14.1 | 84.3 | 12 | |
| Qwen-VLBackbone=Qwen-VL [3]2025.02 | 14.1 | 84.3 | 12 | |
| SigLIP 2Language Model Backbone=Llama-3.2-3B-Instruct2025.12 | 12.8 | — | 9.7 | |
| Pix2StructLanguage Model Backbone=Llama-3.2-3B-Instruct2025.12 | 10.8 | — | 7.5 | |
| DinoV2Language Model Backbone=Llama-3.2-3B-Instruct2025.12 | 9.5 | — | 7.2 | |
| AIMv2Language Model Backbone=Llama-3.2-3B-Instruct2025.12 | 8.8 | — | 6.1 | |
| MiniCPM-V+Backbone=MiniCPM-V [128], DoM Tree=true2025.02 | 6.5 | 81.4 | 5.2 | |
| Fuyu-8B+Backbone=Fuyu-8B [4], DoM Tree=true2025.02 | 3.6 | 83 | 3 | |
| DiTLanguage Model Backbone=Llama-3.2-3B-Instruct2025.12 | 2.1 | — | 1.1 | |
| GPT-4VBackbone=GPT-4V [99], Image=true2025.02 | — | 23.7 | — |