MEGA Hub
New

Wuying-Browser-Agent: Real-World Centric Fundamental Long-Horizon Browser Agents

Published on Aug 18cs.AI

Authors

Do you know AIMAE Team?You can claim authorship or link another user.Do you know Tianxiang Chen?You can claim authorship or link another user.Do you know Yan Cheng?You can claim authorship or link another user.Do you know Zhangye Han?You can claim authorship or link another user.Do you know Xiaowei Li?You can claim authorship or link another user.Do you know Chang Liu?You can claim authorship or link another user.Do you know Cheng Liu?You can claim authorship or link another user.Do you know Zhongqiang Ma?You can claim authorship or link another user.Do you know Long Peng?You can claim authorship or link another user.Do you know Xiaobing Tu?You can claim authorship or link another user.Do you know Yinggui Wang?You can claim authorship or link another user.Do you know Hongliang Wei?You can claim authorship or link another user.Do you know Chen Wu?You can claim authorship or link another user.Do you know Daiping Xin?You can claim authorship or link another user.Do you know Kunyu Zhou?You can claim authorship or link another user.Do you know Pengyang Zhou?You can claim authorship or link another user.Do you know Peiyuan Chen?You can claim authorship or link another user.Do you know Ziyuan Chen?You can claim authorship or link another user.Do you know Yutao Deng?You can claim authorship or link another user.Do you know Chunyu Dong?You can claim authorship or link another user.Do you know Xiangyu Fu?You can claim authorship or link another user.Do you know Yicheng Feng?You can claim authorship or link another user.Do you know Ruian He?You can claim authorship or link another user.Do you know Haochen Li?You can claim authorship or link another user.Do you know Miancan Liu?You can claim authorship or link another user.Do you know Zhengqin Liu?You can claim authorship or link another user.Do you know Wei Peng?You can claim authorship or link another user.Do you know Jinkui Ren?You can claim authorship or link another user.Do you know Haoyu Tan?You can claim authorship or link another user.Do you know Dong Xiao?You can claim authorship or link another user.Do you know Rongkun Xue?You can claim authorship or link another user.Do you know Shujian Yang?You can claim authorship or link another user.Do you know Xianhang Ye?You can claim authorship or link another user.Do you know Ziqi Yuan?You can claim authorship or link another user.Do you know Ziyang Yu?You can claim authorship or link another user.Do you know Linghan Zhang?You can claim authorship or link another user.Do you know Xiantao Zhang?You can claim authorship or link another user.Do you know Xuanpu Zhao?You can claim authorship or link another user.Do you know Yinan Zhao?You can claim authorship or link another user.Do you know Zhenghui Zhao?You can claim authorship or link another user.Do you know Bin Zhu?You can claim authorship or link another user.Do you know Likai Zou?You can claim authorship or link another user.

Abstract

Browser agents perform well on short, clean demonstrations, but real deployment is fundamentally different: agents must sustain dozens of decisions on live websites while recovering from mistakes and navigating complex UIs. We argue that closing this gap requires alignment at every level of the pipeline, including execution, supervision, optimization, and evaluation, rather than scale alone. We present Wuying-Browser-Agent, a unified framework that addresses each of these levels. A structured browser harness provides stable execution primitives and decision-oriented context management. Reflection and UI-specialized Curriculum SFT (RUIC-SFT) explicitly trains on recovery trajectories and complex-UI interactions. Divergence-Aware Online GRPO (DAO-GRPO) improves long-horizon credit assignment through potential-based reward shaping and divergence-aware step weighting. Finally, we introduce BrowserBench, a bilingual real-web benchmark of 350 tasks averaging 37.9 steps, because most existing benchmarks are too short to expose long-horizon failure modes. Wuying-Browser-Agent-27B achieves 80.6\% on WebVoyager, 66.7\% on Online-Mind2Web, and 65.1\% on BrowserBench, establishing a new open-source state of the art on browser-use benchmarks. The same pipeline also transfers beyond browser use, demonstrating strong general agentic ability and reaching an average score of 73.8 on Tau2-Bench, Claw-Eval, and BFCL-v4.

Community

00