Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
video
video
0.13
45.5
label
class label
709 classes
0T0001
0T0001
1T0002
1T0002
2T0003
2T0003
3T0004
3T0004
3T0004
3T0004
4T0005
4T0005
5T0006
5T0006
6T0007
6T0007
6T0007
7T0008
7T0008
8T0009
8T0009
8T0009
9T0010
9T0010
10T0011
10T0011
10T0011
11T0012
11T0012
11T0012
12T0013
12T0013
12T0013
12T0013
13T0014
13T0014
14T0015
14T0015
14T0015
15T0016
15T0016
15T0016
16T0017
16T0017
16T0017
16T0017
16T0017
17T0018
17T0018
17T0018
18T0019
18T0019
18T0019
19T0020
19T0020
19T0020
20T0021
20T0021
20T0021
20T0021
21T0022
21T0022
22T0023
22T0023
22T0023
22T0023
23T0024
23T0024
23T0024
24T0025
24T0025
24T0025
24T0025
25T0031
25T0031
25T0031
25T0031
26T0032
26T0032
26T0032
26T0032
26T0032
26T0032
27T0033
27T0033
27T0033
28T0034
28T0034
28T0034
28T0034
29T0035
29T0035
29T0035
29T0035
30T0037
30T0037
30T0037
30T0037
31T0038
31T0038
End of preview. Expand in Data Studio

OmniGUI: Benchmarking GUI Agents in Omni-Modal Smartphone Environments

OmniGUI is a step-level GUI agent benchmark designed for omni-modal smartphone interaction. At each action step, the agent receives interleaved multimodal observations, including static screenshots, synchronous audio cues, short video clips, and action history, and must predict the next GUI action such as TAP or TYPE.

The benchmark contains 708 expert-demonstrated episodes and 2,572 action steps across 29 real smartphone applications in both Chinese and English. OmniGUI evaluates five core capabilities: Localization, Semantic Understanding, Cross-modal Discrimination, Temporal Reasoning, and Instant Response.

πŸš€ Quick Start

Download the dataset

You can download the dataset using the huggingface-cli:

huggingface-cli download XIAOCHENLIN00zz/OmniGUI \
  --repo-type dataset \
  --local-dir data

After downloading, your local data directory should contain:

  • data/merged_sorted.json
  • data/benchmark/

πŸ› οΈ APP STATISTICS

πŸ“‚ Chinese Applications (ZH)

App Name Ep Stp Loc Sem Cro Tem Ins
Bilibili 29 89 7 5 6 6 5
Douyin 29 97 7 5 6 6 5
Meituan 26 103 6 5 6 5 4
WeChat 25 84 7 5 4 5 4
DiDi 25 92 5 5 5 5 5
Amap 25 64 5 5 5 5 5
JD 25 80 5 5 5 5 5
PDD 25 72 5 5 5 5 5
QQ Music 25 95 5 5 5 5 5
Taobao 21 78 5 5 2 4 5
Weibo 25 106 5 5 5 5 5
RedNote 25 99 5 5 5 5 5
Kuaishou 24 120 4 5 5 6 4
iQIYI 20 56 0 5 0 12 3
Alipay 14 64 5 0 5 4 0
SUBTOTAL 363 1299 76 70 69 83 65

πŸ“‚ English Applications (EN)

App Name Ep Stp Loc Sem Cro Tem Ins
Duolingo 35 133 5 5 5 15 5
GTrans 25 88 5 5 5 5 5
Snapchat 25 104 5 5 5 5 5
Spotify 25 91 5 5 5 5 5
Tasty 25 69 5 5 10 5 0
TED 25 115 5 5 5 5 5
TikTok 25 77 5 5 5 5 5
Vimeo 25 86 5 5 5 5 5
X 25 61 5 5 5 5 5
YouTube 25 123 5 5 5 5 5
RedBull 24 110 5 4 5 5 5
Amazon 22 69 5 5 4 4 4
IMDb 20 95 5 4 4 3 4
Insta 19 52 4 4 4 1 6
SUBTOTAL 345 1273 69 67 72 73 64

πŸ“ˆ FINAL SUMMARY (TOTAL)

Total Ep Stp Loc Sem Cro Tem Ins
ALL APPS 708 2572 145 137 141 156 129

Citation

Feel free to cite the following article if you find OmniGUI helpful:

@misc{henry2026omniguibenchmarkingguiagents,
      title={OmniGUI: Benchmarking GUI Agents in Omni-Modal Smartphone Environments}, 
      author={Felix Henry and Xiaochen Lin and Jiangyou Zhu and Yangfan and Bingqian Zhang and Min Chen and Shiyu Huang},
      year={2026},
      eprint={2605.18758},
      archivePrefix={arXiv},
      primaryClass={cs.HC},
      url={https://arxiv.org/abs/2605.18758}, 
}
Downloads last month
6,814

Paper for OmniGUI/OmniGUI