Phi-Ground Tech Report: Advancing Perception in GUI Grounding
Miaosen Zhang Ziqiang Xu Jialiang Zhu Qi Dai Kai Qiu Yifan Yang · 2025
Read Phi-Ground Tech Report: Advancing Perception in GUI Grounding by Miaosen Zhang Ziqiang Xu Jialiang Zhu Qi Dai Kai Qiu Yifan Yang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-23779.txt.
Description
Miaosen Zhang and a team of co-authors reported in twenty twenty-five on Phi-Ground, a family of models that teach computer-use agents where to click. Such agents act on screens, and grounding — finding the right button or field — decides success. Leading end-to-end models still score under sixty-five percent on hard benchmarks, too low when one misclick can ruin a task.
The report walks through an empirical study of training these models, from data collection to final tuning. The resulting Phi-Ground models top all five grounding benchmarks among models under ten billion parameters in agent settings. In end-to-end settings they also set records on two named benchmarks. The authors share their failures as well as successes, aiming to guide other perception tasks beyond screen grounding.
Catalog record
| CREATOR | Miaosen Zhang Ziqiang Xu Jialiang Zhu Qi Dai Kai Qiu Yifan Yang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 128 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-23779