Add link to paper

#2
by nielsr HF staff - opened
Files changed (1) hide show
  1. README.md +120 -199
README.md CHANGED
@@ -1,199 +1,120 @@
1
- ---
2
- license: apache-2.0
3
- language:
4
- - en
5
- pipeline_tag: image-text-to-text
6
- tags:
7
- - multimodal
8
- - gui
9
- library_name: transformers
10
- ---
11
-
12
-
13
- # UI-TARS-2B-SFT
14
- [UI-TARS-2B-SFT](https://huggingface.co/bytedance-research/UI-TARS-2B-SFT)  | 
15
- [UI-TARS-2B-gguf](https://huggingface.co/bytedance-research/UI-TARS-2B-gguf)  | 
16
- [UI-TARS-7B-SFT](https://huggingface.co/bytedance-research/UI-TARS-7B-SFT)  | 
17
- [**UI-TARS-7B-DPO**](https://huggingface.co/bytedance-research/UI-TARS-7B-DPO)(Recommended)  | 
18
- [UI-TARS-7B-gguf](https://huggingface.co/bytedance-research/UI-TARS-7B-gguf)  | 
19
- [UI-TARS-72B-SFT](https://huggingface.co/bytedance-research/UI-TARS-72B-SFT)  | 
20
- [**UI-TARS-72B-DPO**](https://huggingface.co/bytedance-research/UI-TARS-72B-DPO)(Recommended)
21
- ## Introduction
22
-
23
- UI-TARS is a next-generation native GUI agent model designed to interact seamlessly with graphical user interfaces (GUIs) using human-like perception, reasoning, and action capabilities. Unlike traditional modular frameworks, UI-TARS integrates all key components—perception, reasoning, grounding, and memory—within a single vision-language model (VLM), enabling end-to-end task automation without predefined workflows or manual rules.
24
- <!-- ![Local Image](figures/UI-TARS.png) -->
25
- <p align="center">
26
- <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS-vs-Previous-SOTA.png?raw=true" width="90%"/>
27
- <p>
28
- <p align="center">
29
- <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS.png?raw=true" width="90%"/>
30
- <p>
31
-
32
- <!-- ![Local Image](figures/UI-TARS-vs-Previous-SOTA.png) -->
33
-
34
- This repository contains the model for the paper [UI-TARS: Pioneering Automated GUI Interaction with Native Agents](https://huggingface.co/papers/2501.12326).
35
-
36
- Code: https://github.com/bytedance/UI-TARS
37
-
38
- ## Performance
39
- **Perception Capabilty Evaluation**
40
- | Model | VisualWebBench | WebSRC | SQAshort |
41
- |---------------------------|---------------|---------|----------|
42
- | Qwen2-VL-7B | 73.3 | 81.8 | 84.9 |
43
- | Qwen-VL-Max | 74.1 | 91.1 | 78.6 |
44
- | Gemini-1.5-Pro | 75.4 | 88.9 | 82.2 |
45
- | UIX-Qwen2-7B | 75.9 | 82.9 | 78.8 |
46
- | Claude-3.5-Sonnet | 78.2 | 90.4 | 83.1 |
47
- | GPT-4o | 78.5 | 87.7 | 82.3 |
48
- | **UI-TARS-2B** | 72.9 | 89.2 | 86.4 |
49
- | **UI-TARS-7B** | 79.7 | **93.6** | 87.7 |
50
- | **UI-TARS-72B** | **82.8** | 89.3 | **88.6** |
51
-
52
- **Grounding Capability Evaluation**
53
- - **ScreenSpot Pro**
54
-
55
- | Agent Model | Dev-Text | Dev-Icon | Dev-Avg | Creative-Text | Creative-Icon | Creative-Avg | CAD-Text | CAD-Icon | CAD-Avg | Scientific-Text | Scientific-Icon | Scientific-Avg | Office-Text | Office-Icon | Office-Avg | OS-Text | OS-Icon | OS-Avg | Avg-Text | Avg-Icon | Avg |
56
- |--------------------------|----------|----------|----------|--------------|--------------|--------------|---------|---------|---------|---------------|---------------|---------------|------------|------------|------------|--------|--------|--------|---------|---------|------|
57
- | QwenVL-7B | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.7 | 0.0 | 0.4 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.1 | 0.0 | **0.1** |
58
- | GPT-4o | 1.3 | 0.0 | 0.7 | 1.0 | 0.0 | 0.6 | 2.0 | 0.0 | 1.5 | 2.1 | 0.0 | 1.2 | 1.1 | 0.0 | 0.9 | 0.0 | 0.0 | 0.0 | 1.3 | 0.0 | **0.8** |
59
- | SeeClick | 0.6 | 0.0 | 0.3 | 1.0 | 0.0 | 0.6 | 2.5 | 0.0 | 1.9 | 3.5 | 0.0 | 2.0 | 1.1 | 0.0 | 0.9 | 2.8 | 0.0 | 1.5 | 1.8 | 0.0 | **1.1** |
60
- | Qwen2-VL-7B | 2.6 | 0.0 | 1.3 | 1.5 | 0.0 | 0.9 | 0.5 | 0.0 | 0.4 | 6.3 | 0.0 | 3.5 | 3.4 | 1.9 | 3.0 | 0.9 | 0.0 | 0.5 | 2.5 | 0.2 | **1.6** |
61
- | OS-Atlas-4B | 7.1 | 0.0 | 3.7 | 3.0 | 1.4 | 2.3 | 2.0 | 0.0 | 1.5 | 9.0 | 5.5 | 7.5 | 5.1 | 3.8 | 4.8 | 5.6 | 0.0 | 3.1 | 5.0 | 1.7 | **3.7** |
62
- | ShowUI-2B | 16.9 | 1.4 | 9.4 | 9.1 | 0.0 | 5.3 | 2.5 | 0.0 | 1.9 | 13.2 | 7.3 | 10.6 | 15.3 | 7.5 | 13.5 | 10.3 | 2.2 | 6.6 | 10.8 | 2.6 | **7.7** |
63
- | CogAgent-18B | 14.9 | 0.7 | 8.0 | 9.6 | 0.0 | 5.6 | 7.1 | 3.1 | 6.1 | 22.2 | 1.8 | 13.4 | 13.0 | 0.0 | 10.0 | 5.6 | 0.0 | 3.1 | 12.0 | 0.8 | **7.7** |
64
- | Aria-UI | 16.2 | 0.0 | 8.4 | 23.7 | 2.1 | 14.7 | 7.6 | 1.6 | 6.1 | 27.1 | 6.4 | 18.1 | 20.3 | 1.9 | 16.1 | 4.7 | 0.0 | 2.6 | 17.1 | 2.0 | **11.3** |
65
- | UGround-7B | 26.6 | 2.1 | 14.7 | 27.3 | 2.8 | 17.0 | 14.2 | 1.6 | 11.1 | 31.9 | 2.7 | 19.3 | 31.6 | 11.3 | 27.0 | 17.8 | 0.0 | 9.7 | 25.0 | 2.8 | **16.5** |
66
- | Claude Computer Use | 22.0 | 3.9 | 12.6 | 25.9 | 3.4 | 16.8 | 14.5 | 3.7 | 11.9 | 33.9 | 15.8 | 25.8 | 30.1 | 16.3 | 26.9 | 11.0 | 4.5 | 8.1 | 23.4 | 7.1 | **17.1** |
67
- | OS-Atlas-7B | 33.1 | 1.4 | 17.7 | 28.8 | 2.8 | 17.9 | 12.2 | 4.7 | 10.3 | 37.5 | 7.3 | 24.4 | 33.9 | 5.7 | 27.4 | 27.1 | 4.5 | 16.8 | 28.1 | 4.0 | **18.9** |
68
- | UGround-V1-7B | - | - | 35.5 | - | - | 27.8 | - | - | 13.5 | - | - | 38.8 | - | - | 48.8 | - | - | 26.1 | - | - | **31.1** |
69
- | **UI-TARS-2B** | 47.4 | 4.1 | 26.4 | 42.9 | 6.3 | 27.6 | 17.8 | 4.7 | 14.6 | 56.9 | 17.3 | 39.8 | 50.3 | 17.0 | 42.6 | 21.5 | 5.6 | 14.3 | 39.6 | 8.4 | **27.7** |
70
- | **UI-TARS-7B** | 58.4 | 12.4 | 36.1 | 50.0 | 9.1 | 32.8 | **20.8**| 9.4 | **18.0**| 63.9 | **31.8** | **50.0** | **63.3** | 20.8 | 53.5 | 30.8 | **16.9**| 24.5 | 47.8 | 16.2 | **35.7** |
71
- | **UI-TARS-72B** | **63.0** | **17.3** | **40.8** | **57.1** | **15.4** | **39.6** | 18.8 | **12.5**| 17.2 | **64.6** | 20.9 | 45.7 | **63.3** | **26.4** | **54.8** | **42.1**| 15.7 | **30.1**| **50.9**| **17.5**| **38.1** |
72
-
73
-
74
- - **ScreenSpot**
75
-
76
- | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
77
- |--------|-------------|-------------|-------------|-------------|-------------|---------|---------|
78
- | **Agent Framework** | | | | | | | |
79
- | GPT-4 (SeeClick) | 76.6 | 55.5 | 68.0 | 28.6 | 40.9 | 23.3 | **48.8** |
80
- | GPT-4 (OmniParser) | 93.9 | 57.0 | 91.3 | 63.6 | 81.3 | 51.0 | **73.0** |
81
- | GPT-4 (UGround-7B) | 90.1 | 70.3 | 87.1 | 55.7 | 85.7 | 64.6 | **75.6** |
82
- | GPT-4o (SeeClick) | 81.0 | 59.8 | 69.6 | 33.6 | 43.9 | 26.2 | **52.3** |
83
- | GPT-4o (UGround-7B) | 93.4 | 76.9 | 92.8 | 67.9 | 88.7 | 68.9 | **81.4** |
84
- | **Agent Model** | | | | | | | |
85
- | GPT-4 | 22.6 | 24.5 | 20.2 | 11.8 | 9.2 | 8.8 | **16.2** |
86
- | GPT-4o | 20.2 | 24.9 | 21.1 | 23.6 | 12.2 | 7.8 | **18.3** |
87
- | CogAgent | 67.0 | 24.0 | 74.2 | 20.0 | 70.4 | 28.6 | **47.4** |
88
- | SeeClick | 78.0 | 52.0 | 72.2 | 30.0 | 55.7 | 32.5 | **53.4** |
89
- | Qwen2-VL | 75.5 | 60.7 | 76.3 | 54.3 | 35.2 | 25.7 | **55.3** |
90
- | UGround-7B | 82.8 | 60.3 | 82.5 | 63.6 | 80.4 | 70.4 | **73.3** |
91
- | Aguvis-G-7B | 88.3 | 78.2 | 88.1 | 70.7 | 85.7 | 74.8 | **81.8** |
92
- | OS-Atlas-7B | 93.0 | 72.9 | 91.8 | 62.9 | 90.9 | 74.3 | **82.5** |
93
- | Claude Computer Use | - | - | - | - | - | - | **83.0** |
94
- | Gemini 2.0 (Project Mariner) | - | - | - | - | - | - | **84.0** |
95
- | Aguvis-7B | **95.6** | 77.7 | 93.8 | 67.1 | 88.3 | 75.2 | **84.4** |
96
- | Aguvis-72B | 94.5 | **85.2** | 95.4 | 77.9 | **91.3** | **85.9** | **89.2** |
97
- | **Our Model** | | | | | | | |
98
- | **UI-TARS-2B** | 93.0 | 75.5 | 90.7 | 68.6 | 84.3 | 74.8 | **82.3** |
99
- | **UI-TARS-7B** | 94.5 | **85.2** | **95.9** | 85.7 | 90.0 | 83.5 | **89.5** |
100
- | **UI-TARS-72B** | 94.9 | 82.5 | 89.7 | **88.6** | 88.7 | 85.0 | **88.4** |
101
-
102
-
103
- - **ScreenSpot v2**
104
-
105
- | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
106
- |--------|-------------|-------------|-------------|-------------|-------------|---------|---------|
107
- | **Agent Framework** | | | | | | | |
108
- | GPT-4o (SeeClick) | 85.2 | 58.8 | 79.9 | 37.1 | 72.7 | 30.1 | **63.6** |
109
- | GPT-4o (OS-Atlas-4B) | 95.5 | 75.8 | 79.4 | 49.3 | 90.2 | 66.5 | **79.1** |
110
- | GPT-4o (OS-Atlas-7B) | 96.2 | 83.4 | 89.7 | 69.3 | **94.0** | 79.8 | **87.1** |
111
- | **Agent Model** | | | | | | | |
112
- | SeeClick | 78.4 | 50.7 | 70.1 | 29.3 | 55.2 | 32.5 | **55.1** |
113
- | OS-Atlas-4B | 87.2 | 59.7 | 72.7 | 46.4 | 85.9 | 63.1 | **71.9** |
114
- | OS-Atlas-7B | 95.2 | 75.8 | 90.7 | 63.6 | 90.6 | 77.3 | **84.1** |
115
- | **Our Model** | | | | | | | |
116
- | **UI-TARS-2B** | 95.2 | 79.1 | 90.7 | 68.6 | 87.2 | 78.3 | **84.7** |
117
- | **UI-TARS-7B** | **96.9** | **89.1** | **95.4** | 85.0 | 93.6 | 85.2 | **91.6** |
118
- | **UI-TARS-72B** | 94.8 | 86.3 | 91.2 | **87.9** | 91.5 | **87.7** | **90.3** |
119
-
120
-
121
- **Offline Agent Capability Evaluation**
122
- - **Multimodal Mind2Web**
123
-
124
- | Method | Cross-Task Ele.Acc | Cross-Task Op.F1 | Cross-Task Step SR | Cross-Website Ele.Acc | Cross-Website Op.F1 | Cross-Website Step SR | Cross-Domain Ele.Acc | Cross-Domain Op.F1 | Cross-Domain Step SR |
125
- |--------|----------------------|-------------------|--------------------|----------------------|--------------------|-------------------|--------------------|-------------------|-------------------|
126
- | **Agent Framework** | | | | | | | | | |
127
- | GPT-4o (SeeClick) | 32.1 | - | - | 33.1 | - | - | 33.5 | - | - |
128
- | GPT-4o (UGround) | 47.7 | - | - | 46.0 | - | - | 46.6 | - | - |
129
- | GPT-4o (Aria-UI) | 57.6 | - | - | 57.7 | - | - | 61.4 | - | - |
130
- | GPT-4V (OmniParser) | 42.4 | 87.6 | 39.4 | 41.0 | 84.8 | 36.5 | 45.5 | 85.7 | 42.0 |
131
- | **Agent Model** | | | | | | | | | |
132
- | GPT-4o | 5.7 | 77.2 | 4.3 | 5.7 | 79.0 | 3.9 | 5.5 | 86.4 | 4.5 |
133
- | GPT-4 (SOM) | 29.6 | - | 20.3 | 20.1 | - | 13.9 | 27.0 | - | 23.7 |
134
- | GPT-3.5 (Text-only) | 19.4 | 59.2 | 16.8 | 14.9 | 56.5 | 14.1 | 25.2 | 57.9 | 24.1 |
135
- | GPT-4 (Text-only) | 40.8 | 63.1 | 32.3 | 30.2 | 61.0 | 27.0 | 35.4 | 61.9 | 29.7 |
136
- | Claude | 62.7 | 84.7 | 53.5 | 59.5 | 79.6 | 47.7 | 64.5 | 85.4 | 56.4 |
137
- | Aguvis-7B | 64.2 | 89.8 | 60.4 | 60.7 | 88.1 | 54.6 | 60.4 | 89.2 | 56.6 |
138
- | CogAgent | - | - | 62.3 | - | - | 54.0 | - | - | 59.4 |
139
- | Aguvis-72B | 69.5 | 90.8 | 64.0 | 62.6 | 88.6 | 56.5 | 63.5 | 88.5 | 58.2 |
140
- | **Our Model** | | | | | | | | | |
141
- | **UI-TARS-2B** | 62.3 | 90.0 | 56.3 | 58.5 | 87.2 | 50.8 | 58.8 | 89.6 | 52.3 |
142
- | **UI-TARS-7B** | 73.1 | 92.2 | 67.1 | 68.2 | 90.9 | 61.7 | 66.6 | 90.9 | 60.5 |
143
- | **UI-TARS-72B** | **74.7** | **92.5** | **68.6** | **72.4** | **91.2** | **63.5** | **68.9** | **91.8** | **62.1** |
144
-
145
-
146
- - **Android Control and GUI Odyssey**
147
-
148
- | Agent Models | AndroidControl-Low Type | AndroidControl-Low Grounding | AndroidControl-Low SR | AndroidControl-High Type | AndroidControl-High Grounding | AndroidControl-High SR | GUIOdyssey Type | GUIOdyssey Grounding | GUIOdyssey SR |
149
- |---------------------|----------------------|----------------------|----------------|----------------------|----------------------|----------------|----------------|----------------|----------------|
150
- | Claude | 74.3 | 0.0 | 19.4 | 63.7 | 0.0 | 12.5 | 60.9 | 0.0 | 3.1 |
151
- | GPT-4o | 74.3 | 0.0 | 19.4 | 66.3 | 0.0 | 20.8 | 34.3 | 0.0 | 3.3 |
152
- | SeeClick | 93.0 | 73.4 | 75.0 | 82.9 | 62.9 | 59.1 | 71.0 | 52.4 | 53.9 |
153
- | InternVL-2-4B | 90.9 | 84.1 | 80.1 | 84.1 | 72.7 | 66.7 | 82.1 | 55.5 | 51.5 |
154
- | Qwen2-VL-7B | 91.9 | 86.5 | 82.6 | 83.8 | 77.7 | 69.7 | 83.5 | 65.9 | 60.2 |
155
- | Aria-UI | -- | 87.7 | 67.3 | -- | 43.2 | 10.2 | -- | 86.8 | 36.5 |
156
- | OS-Atlas-4B | 91.9 | 83.8 | 80.6 | 84.7 | 73.8 | 67.5 | 83.5 | 61.4 | 56.4 |
157
- | OS-Atlas-7B | 93.6 | 88.0 | 85.2 | 85.2 | 78.5 | 71.2 | 84.5 | 67.8 | 62.0 |
158
- | Aguvis-7B | -- | -- | 80.5 | -- | -- | 61.5 | -- | -- | -- |
159
- | Aguvis-72B | -- | -- | 84.4 | -- | -- | 66.4 | -- | -- | -- |
160
- | **UI-TARS-2B** | **98.1** | 87.3 | 89.3 | 81.2 | 78.4 | 68.9 | 93.9 | 86.8 | 83.4 |
161
- | **UI-TARS-7B** | 98.0 | 89.3 | 90.8 | 83.7 | 80.5 | 72.5 | 94.6 | 90.1 | 87.0 |
162
- | **UI-TARS-72B** | **98.1** | **89.9** | **91.3** | **85.2** | **81.5** | **74.7** | **95.4** | **91.4** | **88.6** |
163
-
164
- **Online Agent Capability Evaluation**
165
-
166
- | Method | OSWorld (Online) | AndroidWorld (Online) |
167
- |--------|-------------------|------------------|
168
- | **Agent Framework** | | |
169
- | GPT-4o (UGround) | - | 32.8 |
170
- | GPT-4o (Aria-UI) | 15.2 | 44.8 |
171
- | GPT-4o (Aguvis-7B) | 14.8 | 37.1 |
172
- | GPT-4o (Aguvis-72B) | 17.0 | - |
173
- | GPT-4o (OS-Atlas-7B) | 14.6 | - |
174
- | **Agent Model** | | |
175
- | GPT-4o | 5.0 | 34.5 (SoM) |
176
- | Gemini-Pro-1.5 | 5.4 | 22.8 (SoM) |
177
- | Aguvis-72B | 10.3 | 26.1 |
178
- | Claude Computer-Use | 14.9 (15 steps) | 27.9 |
179
- | Claude Computer-Use | 22.0 (50 steps) | - |
180
- | **Our Model** | | |
181
- | **UI-TARS-7B-SFT** | 17.7 (15 steps) | 33.0 |
182
- | **UI-TARS-7B-DPO** | 18.7 (15 steps) | - |
183
- | **UI-TARS-72B-SFT** | 18.8 (15 steps) | **46.6** |
184
- | **UI-TARS-72B-DPO** | **22.7** (15 steps) | - |
185
- | **UI-TARS-72B-DPO** | **24.6** (50 steps) | - |
186
-
187
-
188
- ## Citation
189
- If you find our paper and model useful in your research, feel free to give us a cite.
190
-
191
- ```BibTeX
192
- @article{uitars2025,
193
- author = {Yujia Qin, Yining Ye, Junjie Fang, Haoming Wang, Shihao Liang, Shizuo Tian, Junda Zhang, Jiahao Li, Yunxin Li, Shijue Huang, Wanjun Zhong, Kuanye Li, Jiale Yang, Yu Miao, Woyu Lin, Longxiang Liu, Xu Jiang, Qianli Ma, Jingyu Li, Xiaojun Xiao, Kai Cai, Chuang Li, Yaowei Zheng, Chaolin Jin, Chen Li, Xiao Zhou, Minchao Wang, Haoli Chen, Zhaojian Li, Haihua Yang, Haifeng Liu, Feng Lin, Tao Peng, Xin Liu, Guang Shi},
194
- title = {UI-TARS: Pioneering Automated GUI Interaction with Native Agents},
195
- journal = {arXiv preprint arXiv:2501.12326},
196
- url = {https://github.com/bytedance/UI-TARS},
197
- year = {2025}
198
- }
199
- ```
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - en
5
+ pipeline_tag: image-text-to-text
6
+ tags:
7
+ - multimodal
8
+ - gui
9
+ library_name: transformers
10
+ ---
11
+
12
+ # UI-TARS-2B-SFT
13
+ [UI-TARS-2B-SFT](https://huggingface.co/bytedance-research/UI-TARS-2B-SFT) &nbsp;|&nbsp;
14
+ [UI-TARS-2B-gguf](https://huggingface.co/bytedance-research/UI-TARS-2B-gguf) &nbsp;|&nbsp;
15
+ [UI-TARS-7B-SFT](https://huggingface.co/bytedance-research/UI-TARS-7B-SFT) &nbsp;|&nbsp;
16
+ [**UI-TARS-7B-DPO**](https://huggingface.co/bytedance-research/UI-TARS-7B-DPO)(Recommended) &nbsp;|&nbsp;
17
+ [UI-TARS-7B-gguf](https://huggingface.co/bytedance-research/UI-TARS-7B-gguf) &nbsp;|&nbsp;
18
+ [UI-TARS-72B-SFT](https://huggingface.co/bytedance-research/UI-TARS-72B-SFT) &nbsp;|&nbsp;
19
+ [**UI-TARS-72B-DPO**](https://huggingface.co/bytedance-research/UI-TARS-72B-DPO)(Recommended)
20
+ ## Introduction
21
+
22
+ UI-TARS is a next-generation native GUI agent model designed to interact seamlessly with graphical user interfaces (GUIs) using human-like perception, reasoning, and action capabilities. Unlike traditional modular frameworks, UI-TARS integrates all key components—perception, reasoning, grounding, and memory—within a single vision-language model (VLM), enabling end-to-end task automation without predefined workflows or manual rules.
23
+ <!-- ![Local Image](figures/UI-TARS.png) -->
24
+ <p align="center">
25
+ <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS-vs-Previous-SOTA.png?raw=true" width="90%"/>
26
+ <p>
27
+ <p align="center">
28
+ <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS.png?raw=true" width="90%"/>
29
+ <p>
30
+
31
+ <!-- ![Local Image](figures/UI-TARS-vs-Previous-SOTA.png) -->
32
+
33
+ This repository contains the model for the paper [UI-TARS: Pioneering Automated GUI Interaction with Native Agents](https://huggingface.co/papers/2501.12326).
34
+
35
+ Code: https://github.com/bytedance/UI-TARS
36
+
37
+ ## Performance
38
+ **Perception Capabilty Evaluation**
39
+ | Model | VisualWebBench | WebSRC | SQAshort |
40
+ |---------------------------|---------------|---------|----------|
41
+ | Qwen2-VL-7B | 73.3 | 81.8 | 84.9 |
42
+ | Qwen-VL-Max | 74.1 | 91.1 | 78.6 |
43
+ | Gemini-1.5-Pro | 75.4 | 88.9 | 82.2 |
44
+ | UIX-Qwen2-7B | 75.9 | 82.9 | 78.8 |
45
+ | Claude-3.5-Sonnet | 78.2 | 90.4 | 83.1 |
46
+ | GPT-4o | 78.5 | 87.7 | 82.3 |
47
+ | **UI-TARS-2B** | 72.9 | 89.2 | 86.4 |
48
+ | **UI-TARS-7B** | 79.7 | **93.6** | 87.7 |
49
+ | **UI-TARS-72B** | **82.8** | 89.3 | **88.6** |
50
+
51
+ **Grounding Capability Evaluation**
52
+ - **ScreenSpot Pro**
53
+
54
+ | Agent Model | Dev-Text | Dev-Icon | Dev-Avg | Creative-Text | Creative-Icon | Creative-Avg | CAD-Text | CAD-Icon | CAD-Avg | Scientific-Text | Scientific-Icon | Scientific-Avg | Office-Text | Office-Icon | Office-Avg | OS-Text | OS-Icon | OS-Avg | Avg-Text | Avg-Icon | Avg |
55
+ |--------------------------|----------|----------|----------|--------------|--------------|--------------|---------|---------|---------|---------------|---------------|---------------|------------|------------|------------|--------|--------|--------|---------|---------|------|
56
+ | QwenVL-7B | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.7 | 0.0 | 0.4 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.1 | 0.0 | **0.1** |
57
+ | GPT-4o | 1.3 | 0.0 | 0.7 | 1.0 | 0.0 | 0.6 | 2.0 | 0.0 | 1.5 | 2.1 | 0.0 | 1.2 | 1.1 | 0.0 | 0.9 | 0.0 | 0.0 | 0.0 | 1.3 | 0.0 | **0.8** |
58
+ | SeeClick | 0.6 | 0.0 | 0.3 | 1.0 | 0.0 | 0.6 | 2.5 | 0.0 | 1.9 | 3.5 | 0.0 | 2.0 | 1.1 | 0.0 | 0.9 | 2.8 | 0.0 | 1.5 | 1.8 | 0.0 | **1.1** |
59
+ | Qwen2-VL-7B | 2.6 | 0.0 | 1.3 | 1.5 | 0.0 | 0.9 | 0.5 | 0.0 | 0.4 | 6.3 | 0.0 | 3.5 | 3.4 | 1.9 | 3.0 | 0.9 | 0.0 | 0.5 | 2.5 | 0.2 | **1.6** |
60
+ | OS-Atlas-4B | 7.1 | 0.0 | 3.7 | 3.0 | 1.4 | 2.3 | 2.0 | 0.0 | 1.5 | 9.0 | 5.5 | 7.5 | 5.1 | 3.8 | 4.8 | 5.6 | 0.0 | 3.1 | 5.0 | 1.7 | **3.7** |
61
+ | ShowUI-2B | 16.9 | 1.4 | 9.4 | 9.1 | 0.0 | 5.3 | 2.5 | 0.0 | 1.9 | 13.2 | 7.3 | 10.6 | 15.3 | 7.5 | 13.5 | 10.3 | 2.2 | 6.6 | 10.8 | 2.6 | **7.7** |
62
+ | CogAgent-18B | 14.9 | 0.7 | 8.0 | 9.6 | 0.0 | 5.6 | 7.1 | 3.1 | 6.1 | 22.2 | 1.8 | 13.4 | 13.0 | 0.0 | 10.0 | 5.6 | 0.0 | 3.1 | 12.0 | 0.8 | **7.7** |
63
+ | Aria-UI | 16.2 | 0.0 | 8.4 | 23.7 | 2.1 | 14.7 | 7.6 | 1.6 | 6.1 | 27.1 | 6.4 | 18.1 | 20.3 | 1.9 | 16.1 | 4.7 | 0.0 | 2.6 | 17.1 | 2.0 | **11.3** |
64
+ | UGround-7B | 26.6 | 2.1 | 14.7 | 27.3 | 2.8 | 17.0 | 14.2 | 1.6 | 11.1 | 31.9 | 2.7 | 19.3 | 31.6 | 11.3 | 27.0 | 17.8 | 0.0 | 9.7 | 25.0 | 2.8 | **16.5** |
65
+ | Claude Computer Use | 22.0 | 3.9 | 12.6 | 25.9 | 3.4 | 16.8 | 14.5 | 3.7 | 11.9 | 33.9 | 15.8 | 25.8 | 30.1 | 16.3 | 26.9 | 11.0 | 4.5 | 8.1 | 23.4 | 7.1 | **17.1** |
66
+ | OS-Atlas-7B | 33.1 | 1.4 | 17.7 | 28.8 | 2.8 | 17.9 | 12.2 | 4.7 | 10.3 | 37.5 | 7.3 | 24.4 | 33.9 | 5.7 | 27.4 | 27.1 | 4.5 | 16.8 | 28.1 | 4.0 | **18.9** |
67
+ | UGround-V1-7B | - | - | 35.5 | - | - | 27.8 | - | - | 13.5 | - | - | 38.8 | - | - | 48.8 | - | - | 26.1 | - | - | **31.1** |
68
+ | **UI-TARS-2B** | 47.4 | 4.1 | 26.4 | 42.9 | 6.3 | 27.6 | 17.8 | 4.7 | 14.6 | 56.9 | 17.3 | 39.8 | 50.3 | 17.0 | 42.6 | 21.5 | 5.6 | 14.3 | 39.6 | 8.4 | **27.7** |
69
+ | **UI-TARS-7B** | 58.4 | 12.4 | 36.1 | 50.0 | 9.1 | 32.8 | **20.8**| 9.4 | **18.0**| 63.9 | **31.8** | **50.0** | **63.3** | 20.8 | 53.5 | 30.8 | **16.9**| 24.5 | 47.8 | 16.2 | **35.7** |
70
+ | **UI-TARS-72B** | **63.0** | **17.3** | **40.8** | **57.1** | **15.4** | **39.6** | 18.8 | **12.5**| 17.2 | **64.6** | 20.9 | 45.7 | **63.3** | **26.4** | **54.8** | **42.1**| 15.7 | **30.1**| **50.9**| **17.5**| **38.1** |
71
+
72
+
73
+ - **ScreenSpot v2**
74
+
75
+ | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
76
+ |--------|-------------|-------------|-------------|-------------|-------------|---------|---------|
77
+ | **Agent Framework** | | | | | | | |
78
+ | GPT-4o (SeeClick) | 85.2 | 58.8 | 79.9 | 37.1 | 72.7 | 30.1 | **63.6** |
79
+ | GPT-4o (OS-Atlas-4B) | 95.5 | 75.8 | 79.4 | 49.3 | 90.2 | 66.5 | **79.1** |
80
+ | GPT-4o (OS-Atlas-7B) | 96.2 | 83.4 | 89.7 | 69.3 | **94.0** | 79.8 | **87.1** |
81
+ | **Agent Model** | | | | | | | |
82
+ | SeeClick | 78.4 | 50.7 | 70.1 | 29.3 | 55.2 | 32.5 | **55.1** |
83
+ | OS-Atlas-4B | 87.2 | 59.7 | 72.7 | 46.4 | 85.9 | 63.1 | **71.9** |
84
+ | OS-Atlas-7B | 95.2 | 75.8 | 90.7 | 63.6 | 90.6 | 77.3 | **84.1** |
85
+ | **Our Model** | | | | | | | |
86
+ | **UI-TARS-2B** | 95.2 | 79.1 | 90.7 | 68.6 | 87.2 | 78.3 | **84.7** |
87
+ | **UI-TARS-7B** | **96.9** | **89.1** | **95.4** | 85.0 | 93.6 | 85.2 | **91.6** |
88
+ | **UI-TARS-72B** | 94.8 | 86.3 | 91.2 | **87.9** | 91.5 | **87.7** | **90.3** |
89
+
90
+
91
+ **Online Agent Capability Evaluation**
92
+
93
+ | Method | OSWorld (Online) | AndroidWorld (Online) |
94
+ |--------|-------------------|------------------|
95
+ | **Agent Framework** | | |
96
+ | GPT-4o (UGround) | - | 32.8 |
97
+ | GPT-4o (Aria-UI) | 15.2 | 44.8 |
98
+ | GPT-4o (Aguvis-7B) | 14.8 | 37.1 |
99
+ | GPT-4o (Aguvis-72B) | 17.0 | - |
100
+ | GPT-4o (OS-Atlas-7B) | 14.6 | - |
101
+ | **Agent Model** | | |
102
+ | GPT-4o | 5.0 | 34.5 (SoM) |
103
+ | Gemini-Pro-1.5 | 5.4 | 22.8 (SoM) |
104
+ | Aguvis-72B | 10.3 | 26.1 |
105
+ | Claude Computer-Use | 14.9 (15 steps) | 27.9 |
106
+ | Claude Computer-Use | 22.0 (50 steps) | - |
107
+ | **Our Model** | | |
108
+ | **UI-TARS-7B-SFT** | 17.7 (15 steps) | 33.0 |
109
+ | **UI-TARS-7B-DPO** | 18.7 (15 steps) | - |
110
+ | **UI-TARS-72B-SFT** | 18.8 (15 steps) | **46.6** |
111
+ | **UI-TARS-72B-DPO** | **22.7** (15 steps) | - |
112
+ | **UI-TARS-72B-DPO** | **24.6** (50 steps) | - |
113
+
114
+
115
+ ## How to use
116
+ The usage will be the same as LLaVA with a few additional tokens.
117
+ Please refer to https://github.com/haotian-liu/LLaVA
118
+
119
+ ## License
120
+ UI-TARS is licensed under the Apache License 2.0.