Hugging Face's logo

bytedance-research
/
UI-TARS-7B-SFT

Image-Text-to-Text
Transformers
Safetensors
English
qwen2_vl
multimodal
gui
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
3
MingComplex commited on
Commit
4d50b8d
·
1 Parent(s): d889b39

update README.md

Browse files
Files changed (1) hide show
  1. README.md +199 -0
README.md CHANGED
@@ -1,3 +1,202 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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-7B-SFT
14
+
15
+ ## Introduction
16
+
17
+ 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.
18
+ ![Local Image](figures/UI-TARS.png)
19
+ ![Local Image](figures/UI-TARS-vs-Previous-SOTA.png)
20
+
21
+ ## Core Features
22
+ ### Perception
23
+ - **Comprehensive GUI Understanding**: Processes multimodal inputs (text, images, interactions) to build a coherent understanding of interfaces.
24
+ - **Real-Time Interaction**: Continuously monitors dynamic GUIs and responds accurately to changes in real-time.
25
+
26
+ ### Action
27
+ - **Unified Action Space**: Standardized action definitions across platforms (desktop, mobile, and web).
28
+ - **Platform-Specific Actions**: Supports additional actions like hotkeys, long press, and platform-specific gestures.
29
+
30
+ ### Reasoning
31
+ - **System 1 & System 2 Reasoning**: Combines fast, intuitive responses with deliberate, high-level planning for complex tasks.
32
+ - **Task Decomposition & Reflection**: Supports multi-step planning, reflection, and error correction for robust task execution.
33
+
34
+ ### Memory
35
+ - **Short-Term Memory**: Captures task-specific context for situational awareness.
36
+ - **Long-Term Memory**: Retains historical interactions and knowledge for improved decision-making.
37
+
38
+ ## Capabilities
39
+ - **Cross-Platform Interaction**: Supports desktop, mobile, and web environments with a unified action framework.
40
+ - **Multi-Step Task Execution**: Trained to handle complex tasks through multi-step trajectories and reasoning.
41
+ - **Learning from Synthetic and Real Data**: Combines large-scale annotated and synthetic datasets for improved generalization and robustness.
42
+
43
+ ## Performance
44
+ **Perception Capabilty Evaluation**
45
+ | Model | VisualWebBench | WebSRC | SQAshort |
46
+ |---------------------------|---------------|---------|----------|
47
+ | Qwen2-VL-7B | 73.3 | 81.8 | 84.9 |
48
+ | Qwen-VL-Max | 74.1 | 91.1 | 78.6 |
49
+ | Gemini-1.5-Pro | 75.4 | 88.9 | 82.2 |
50
+ | UIX-Qwen2-7B | 75.9 | 82.9 | 78.8 |
51
+ | Claude-3.5-Sonnet | 78.2 | 90.4 | 83.1 |
52
+ | GPT-4o | 78.5 | 87.7 | 82.3 |
53
+ | **UI-TARS-2B** | 72.9 | 89.2 | 86.4 |
54
+ | **UI-TARS-7B** | 79.7 | **93.6** | 87.7 |
55
+ | **UI-TARS-72B** | **82.8** | 89.3 | **88.6** |
56
+
57
+ **Grounding Capability Evaluation**
58
+ - **ScreenSpot Pro**
59
+
60
+ | 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 |
61
+ |--------------------------|----------|----------|----------|--------------|--------------|--------------|---------|---------|---------|---------------|---------------|---------------|------------|------------|------------|--------|--------|--------|---------|---------|------|
62
+ | 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** |
63
+ | 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** |
64
+ | 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** |
65
+ | 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** |
66
+ | 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** |
67
+ | 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** |
68
+ | 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** |
69
+ | 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** |
70
+ | 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** |
71
+ | 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** |
72
+ | 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** |
73
+ | UGround-V1-7B | - | - | 35.5 | - | - | 27.8 | - | - | 13.5 | - | - | 38.8 | - | - | 48.8 | - | - | 26.1 | - | - | **31.1** |
74
+ | **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** |
75
+ | **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** |
76
+ | **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** |
77
+
78
+
79
+ - **ScreenSpot**
80
+
81
+ | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
82
+ |--------|-------------|-------------|-------------|-------------|-------------|---------|---------|
83
+ | **Agent Framework** | | | | | | | |
84
+ | GPT-4 (SeeClick) | 76.6 | 55.5 | 68.0 | 28.6 | 40.9 | 23.3 | **48.8** |
85
+ | GPT-4 (OmniParser) | 93.9 | 57.0 | 91.3 | 63.6 | 81.3 | 51.0 | **73.0** |
86
+ | GPT-4 (UGround-7B) | 90.1 | 70.3 | 87.1 | 55.7 | 85.7 | 64.6 | **75.6** |
87
+ | GPT-4o (SeeClick) | 81.0 | 59.8 | 69.6 | 33.6 | 43.9 | 26.2 | **52.3** |
88
+ | GPT-4o (UGround-7B) | 93.4 | 76.9 | 92.8 | 67.9 | 88.7 | 68.9 | **81.4** |
89
+ | **Agent Model** | | | | | | | |
90
+ | GPT-4 | 22.6 | 24.5 | 20.2 | 11.8 | 9.2 | 8.8 | **16.2** |
91
+ | GPT-4o | 20.2 | 24.9 | 21.1 | 23.6 | 12.2 | 7.8 | **18.3** |
92
+ | CogAgent | 67.0 | 24.0 | 74.2 | 20.0 | 70.4 | 28.6 | **47.4** |
93
+ | SeeClick | 78.0 | 52.0 | 72.2 | 30.0 | 55.7 | 32.5 | **53.4** |
94
+ | Qwen2-VL | 75.5 | 60.7 | 76.3 | 54.3 | 35.2 | 25.7 | **55.3** |
95
+ | UGround-7B | 82.8 | 60.3 | 82.5 | 63.6 | 80.4 | 70.4 | **73.3** |
96
+ | Aguvis-G-7B | 88.3 | 78.2 | 88.1 | 70.7 | 85.7 | 74.8 | **81.8** |
97
+ | OS-Atlas-7B | 93.0 | 72.9 | 91.8 | 62.9 | 90.9 | 74.3 | **82.5** |
98
+ | Claude Computer Use | - | - | - | - | - | - | **83.0** |
99
+ | Gemini 2.0 (Project Mariner) | - | - | - | - | - | - | **84.0** |
100
+ | Aguvis-7B | **95.6** | 77.7 | 93.8 | 67.1 | 88.3 | 75.2 | **84.4** |
101
+ | Aguvis-72B | 94.5 | **85.2** | 95.4 | 77.9 | **91.3** | **85.9** | **89.2** |
102
+ | **Our Model** | | | | | | | |
103
+ | **UI-TARS-2B** | 93.0 | 75.5 | 90.7 | 68.6 | 84.3 | 74.8 | **82.3** |
104
+ | **UI-TARS-7B** | 94.5 | **85.2** | **95.9** | 85.7 | 90.0 | 83.5 | **89.5** |
105
+ | **UI-TARS-72B** | 94.9 | 82.5 | 89.7 | **88.6** | 88.7 | 85.0 | **88.4** |
106
+
107
+
108
+ - **ScreenSpot v2**
109
+
110
+ | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
111
+ |--------|-------------|-------------|-------------|-------------|-------------|---------|---------|
112
+ | **Agent Framework** | | | | | | | |
113
+ | GPT-4o (SeeClick) | 85.2 | 58.8 | 79.9 | 37.1 | 72.7 | 30.1 | **63.6** |
114
+ | GPT-4o (OS-Atlas-4B) | 95.5 | 75.8 | 79.4 | 49.3 | 90.2 | 66.5 | **79.1** |
115
+ | GPT-4o (OS-Atlas-7B) | 96.2 | 83.4 | 89.7 | 69.3 | **94.0** | 79.8 | **87.1** |
116
+ | **Agent Model** | | | | | | | |
117
+ | SeeClick | 78.4 | 50.7 | 70.1 | 29.3 | 55.2 | 32.5 | **55.1** |
118
+ | OS-Atlas-4B | 87.2 | 59.7 | 72.7 | 46.4 | 85.9 | 63.1 | **71.9** |
119
+ | OS-Atlas-7B | 95.2 | 75.8 | 90.7 | 63.6 | 90.6 | 77.3 | **84.1** |
120
+ | **Our Model** | | | | | | | |
121
+ | **UI-TARS-2B** | 95.2 | 79.1 | 90.7 | 68.6 | 87.2 | 78.3 | **84.7** |
122
+ | **UI-TARS-7B** | **96.9** | **89.1** | **95.4** | 85.0 | 93.6 | 85.2 | **91.6** |
123
+ | **UI-TARS-72B** | 94.8 | 86.3 | 91.2 | **87.9** | 91.5 | **87.7** | **90.3** |
124
+
125
+
126
+ **Offline Agent Capability Evaluation**
127
+ - **Multimodal Mind2Web**
128
+
129
+ | 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 |
130
+ |--------|----------------------|-------------------|--------------------|----------------------|--------------------|-------------------|--------------------|-------------------|-------------------|
131
+ | **Agent Framework** | | | | | | | | | |
132
+ | GPT-4o (SeeClick) | 32.1 | - | - | 33.1 | - | - | 33.5 | - | - |
133
+ | GPT-4o (UGround) | 47.7 | - | - | 46.0 | - | - | 46.6 | - | - |
134
+ | GPT-4o (Aria-UI) | 57.6 | - | - | 57.7 | - | - | 61.4 | - | - |
135
+ | GPT-4V (OmniParser) | 42.4 | 87.6 | 39.4 | 41.0 | 84.8 | 36.5 | 45.5 | 85.7 | 42.0 |
136
+ | **Agent Model** | | | | | | | | | |
137
+ | GPT-4o | 5.7 | 77.2 | 4.3 | 5.7 | 79.0 | 3.9 | 5.5 | 86.4 | 4.5 |
138
+ | GPT-4 (SOM) | 29.6 | - | 20.3 | 20.1 | - | 13.9 | 27.0 | - | 23.7 |
139
+ | GPT-3.5 (Text-only) | 19.4 | 59.2 | 16.8 | 14.9 | 56.5 | 14.1 | 25.2 | 57.9 | 24.1 |
140
+ | GPT-4 (Text-only) | 40.8 | 63.1 | 32.3 | 30.2 | 61.0 | 27.0 | 35.4 | 61.9 | 29.7 |
141
+ | Claude | 62.7 | 84.7 | 53.5 | 59.5 | 79.6 | 47.7 | 64.5 | 85.4 | 56.4 |
142
+ | Aguvis-7B | 64.2 | 89.8 | 60.4 | 60.7 | 88.1 | 54.6 | 60.4 | 89.2 | 56.6 |
143
+ | CogAgent | - | - | 62.3 | - | - | 54.0 | - | - | 59.4 |
144
+ | Aguvis-72B | 69.5 | 90.8 | 64.0 | 62.6 | 88.6 | 56.5 | 63.5 | 88.5 | 58.2 |
145
+ | **Our Model** | | | | | | | | | |
146
+ | **UI-TARS-2B** | 62.3 | 90.0 | 56.3 | 58.5 | 87.2 | 50.8 | 58.8 | 89.6 | 52.3 |
147
+ | **UI-TARS-7B** | 73.1 | 92.2 | 67.1 | 68.2 | 90.9 | 61.7 | 66.6 | 90.9 | 60.5 |
148
+ | **UI-TARS-72B** | **74.7** | **92.5** | **68.6** | **72.4** | **91.2** | **63.5** | **68.9** | **91.8** | **62.1** |
149
+
150
+
151
+ - **Android Control and GUI Odyssey**
152
+
153
+ | 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 |
154
+ |---------------------|----------------------|----------------------|----------------|----------------------|----------------------|----------------|----------------|----------------|----------------|
155
+ | Claude | 74.3 | 0.0 | 19.4 | 63.7 | 0.0 | 12.5 | 60.9 | 0.0 | 3.1 |
156
+ | GPT-4o | 74.3 | 0.0 | 19.4 | 66.3 | 0.0 | 20.8 | 34.3 | 0.0 | 3.3 |
157
+ | SeeClick | 93.0 | 73.4 | 75.0 | 82.9 | 62.9 | 59.1 | 71.0 | 52.4 | 53.9 |
158
+ | InternVL-2-4B | 90.9 | 84.1 | 80.1 | 84.1 | 72.7 | 66.7 | 82.1 | 55.5 | 51.5 |
159
+ | Qwen2-VL-7B | 91.9 | 86.5 | 82.6 | 83.8 | 77.7 | 69.7 | 83.5 | 65.9 | 60.2 |
160
+ | Aria-UI | -- | 87.7 | 67.3 | -- | 43.2 | 10.2 | -- | 86.8 | 36.5 |
161
+ | OS-Atlas-4B | 91.9 | 83.8 | 80.6 | 84.7 | 73.8 | 67.5 | 83.5 | 61.4 | 56.4 |
162
+ | OS-Atlas-7B | 93.6 | 88.0 | 85.2 | 85.2 | 78.5 | 71.2 | 84.5 | 67.8 | 62.0 |
163
+ | Aguvis-7B | -- | -- | 80.5 | -- | -- | 61.5 | -- | -- | -- |
164
+ | Aguvis-72B | -- | -- | 84.4 | -- | -- | 66.4 | -- | -- | -- |
165
+ | **UI-TARS-2B** | **98.1** | 87.3 | 89.3 | 81.2 | 78.4 | 68.9 | 93.9 | 86.8 | 83.4 |
166
+ | **UI-TARS-7B** | 98.0 | 89.3 | 90.8 | 83.7 | 80.5 | 72.5 | 94.6 | 90.1 | 87.0 |
167
+ | **UI-TARS-72B** | **98.1** | **89.9** | **91.3** | **85.2** | **81.5** | **74.7** | **95.4** | **91.4** | **88.6** |
168
+
169
+ **Online Agent Capability Evaluation**
170
+
171
+ | Method | OSWorld (Online) | AndroidWorld (Online) |
172
+ |--------|-------------------|------------------|
173
+ | **Agent Framework** | | |
174
+ | GPT-4o (UGround) | - | 32.8 |
175
+ | GPT-4o (Aria-UI) | 15.2 | 44.8 |
176
+ | GPT-4o (Aguvis-7B) | 14.8 | 37.1 |
177
+ | GPT-4o (Aguvis-72B) | 17.0 | - |
178
+ | GPT-4o (OS-Atlas-7B) | 14.6 | - |
179
+ | **Agent Model** | | |
180
+ | GPT-4o | 5.0 | 34.5 (SoM) |
181
+ | Gemini-Pro-1.5 | 5.4 | 22.8 (SoM) |
182
+ | Aguvis-72B | 10.3 | 26.1 |
183
+ | Claude Computer-Use | 14.9 (15 steps) | 27.9 |
184
+ | Claude Computer-Use | 22.0 (50 steps) | - |
185
+ | **Our Model** | | |
186
+ | **UI-TARS-7B-SFT** | 17.7 (15 steps) | 33.0 |
187
+ | **UI-TARS-7B-DPO** | 18.7 (15 steps) | - |
188
+ | **UI-TARS-72B-SFT** | 18.8 (15 steps) | **46.6** |
189
+ | **UI-TARS-72B-DPO** | **22.7** (15 steps) | - |
190
+
191
+
192
+ ## Citation
193
+ If you find our paper and code useful in your research, please consider giving a star :star: and citation :pencil:
194
+
195
+ ```BibTeX
196
+ @article{uitars2025,
197
+ 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, Xin Jin, Chen Li, Xiao Zhou, Minchao Wang, Haoli Chen, Zhaojian Li, Haihua Yang, Haifeng Liu, Feng Lin, Tao Peng, Xin Liu, Guang Shi},
198
+ title = {UI-TARS: An End-to-End Framework for Autonomous GUI Agents with System-2 Reasoning and Iterative Reflection Tuning},
199
+ url = {https://github.com/bytedance/UI-TARS},
200
+ year = {2025}
201
+ }
202
+ ```