UI-TARS-2 Technical Report: Advancing GUI Agent with Multi-Turn Reinforcement Learning
Abstract
The development of autonomous agents for graphical user interfaces (GUIs) presents major challenges in artificial intelligence. While recent advances in native agent models have shown promise by unifying perception, reasoning, action, and memory through end-to-end learning, open problems remain in data scalability, multi-turn reinforcement learning (RL), the limitations of GUI-only operation, and environment stability. In this technical report, we present UI-TARS-2, a native GUI-centered agent model that addresses these challenges through a systematic training methodology: a data flywheel for scalable data generation, a stabilized multi-turn RL framework, a hybrid GUI environment that integrates file systems and terminals, and a unified sandbox platform for large-scale rollouts. Empirical evaluation demonstrates that UI-TARS-2 achieves significant improvements over its predecessor UI-TARS-1.5. On GUI benchmarks, it reaches 88.2 on Online-Mind2Web, 47.5 on OSWorld, 50.6 on WindowsAgentArena, and 73.3 on AndroidWorld, outperforming strong baselines such as Claude and OpenAI agents. In game environments, it attains a mean normalized score of 59.8 across a 15-game suite-roughly 60% of human-level performance-and remains competitive with frontier proprietary models (e.g., OpenAI o3) on LMGame-Bench. Additionally, the model can generalize to long-horizon information-seeking tasks and software engineering benchmarks, highlighting its robustness across diverse agent tasks. Detailed analyses of training dynamics further provide insights into achieving stability and efficiency in large-scale agent RL. These results underscore UI-TARS-2's potential to advance the state of GUI agents and exhibit strong generalization to real-world interactive scenarios.
Cite
@article{arxiv.2509.02544,
title = {UI-TARS-2 Technical Report: Advancing GUI Agent with Multi-Turn Reinforcement Learning},
author = {Haoming Wang and Haoyang Zou and Huatong Song and Jiazhan Feng and Junjie Fang and Junting Lu and Longxiang Liu and Qinyu Luo and Shihao Liang and Shijue Huang and Wanjun Zhong and Yining Ye and Yujia Qin and Yuwen Xiong and Yuxin Song and Zhiyong Wu and Aoyan Li and Bo Li and Chen Dun and Chong Liu and Daoguang Zan and Fuxing Leng and Hanbin Wang and Hao Yu and Haobin Chen and Hongyi Guo and Jing Su and Jingjia Huang and Kai Shen and Kaiyu Shi and Lin Yan and Peiyao Zhao and Pengfei Liu and Qinghao Ye and Renjie Zheng and Shulin Xin and Wayne Xin Zhao and Wen Heng and Wenhao Huang and Wenqian Wang and Xiaobo Qin and Yi Lin and Youbin Wu and Zehui Chen and Zihao Wang and Baoquan Zhong and Xinchun Zhang and Xujing Li and Yuanfan Li and Zhongkai Zhao and Chengquan Jiang and Faming Wu and Haotian Zhou and Jinlin Pang and Li Han and Qi Liu and Qianli Ma and Siyao Liu and Songhua Cai and Wenqi Fu and Xin Liu and Yaohui Wang and Zhi Zhang and Bo Zhou and Guoliang Li and Jiajun Shi and Jiale Yang and Jie Tang and Li Li and Qihua Han and Taoran Lu and Woyu Lin and Xiaokang Tong and Xinyao Li and Yichi Zhang and Yu Miao and Zhengxuan Jiang and Zili Li and Ziyuan Zhao and Chenxin Li and Dehua Ma and Feng Lin and Ge Zhang and Haihua Yang and Hangyu Guo and Hongda Zhu and Jiaheng Liu and Junda Du and Kai Cai and Kuanye Li and Lichen Yuan and Meilan Han and Minchao Wang and Shuyue Guo and Tianhao Cheng and Xiaobo Ma and Xiaojun Xiao and Xiaolong Huang and Xinjie Chen and Yidi Du and Yilin Chen and Yiwen Wang and Zhaojian Li and Zhenzhu Yang and Zhiyuan Zeng and Chaolin Jin and Chen Li and Hao Chen and Haoli Chen and Jian Chen and Qinghao Zhao and Guang Shi},
journal= {arXiv preprint arXiv:2509.02544},
year = {2025}
}