AgentCPM‑GUI: Building Mobile‑Use Agents with Reinforcement Fine‑Tuning

Zhong Zhang, Yaxi Lu, Yikun Fu, Yupeng Huo, Shenzhi Yang, Yesai Wu, Han Si, Xin Cong, Haotian Chen, Yankai Lin, Jie Xie, Wei Zhou, Wang Xu, Yuanheng Zhang, Zhou Su, Zhongwu Zhai, Xiaoming Liu, Yudong Mei, Jianming Xu, Hongyan Tian, Chongyi Wang, Chi Chen, Yuan Yao, Zhiyuan Liu, Maosong Sun

🏛 Institutions: Tsinghua, Renmin University of China, ModelBest
📅 Date: June 2, 2025
📑 Publisher: EMNLP 2025 System Demonstrations
💻 Env: Mobile
🔑 Keywords: model benchmark reinforcement learning GRPO grounding-aware pre-training CAGUI AgentCPM-GUI

TLDR

AgentCPM-GUI is an 8B mobile GUI model aimed at robust on-device interaction, especially for Chinese and English interfaces. It combines grounding-aware pre-training, supervised trajectory imitation, and GRPO-based reinforcement fine-tuning, and reports strong results on five public benchmarks plus the newly proposed Chinese benchmark CAGUI.

Open paper Edit on GitHub Report issue