UI-Venus-2 Technical Report

Abstract: Multimodal GUI agents have emerged as a promising paradigm for digital task automation, yet transitioning from benchmark-oriented models to dependable real-world applications remains challenging due to limited environment coverage, brittle task construction, and unreliable reward verification.

摘要： 多模态图形用户界面（GUI）智能体已成为数字任务自动化领域一种极具前景的范式。然而，由于环境覆盖范围有限、任务构建脆弱以及奖励验证不可靠，从基准测试导向的模型向可靠的现实世界应用转型仍然充满挑战。

In this work, we present UI-Venus-2, a general-purpose foundation GUI agent designed to operate across mobile, web, and desktop environments through a unified closed-loop reasoning-action framework.

在这项工作中，我们提出了 UI-Venus-2，这是一个通用的基础 GUI 智能体，旨在通过统一的闭环推理-行动框架，在移动端、网页端和桌面端环境中运行。

To bridge the gap toward practical deployment, we jointly scale three critical dimensions: (1) Environments, expanding coverage to more than 170 multilingual mobile apps and native desktop operating systems; (2) Tasks, employing a deep-research pipeline for function-grounded instruction generation; and (3) Verification, adopting trace-level and sample-level evaluators with visual keypoints and multi-model voting to ensure reliable RL signals for training.

为了弥合通往实际部署的差距，我们共同扩展了三个关键维度：(1) 环境，将覆盖范围扩大到超过 170 个多语言移动应用程序和原生桌面操作系统；(2) 任务，采用深度研究流水线进行基于功能基础的指令生成；(3) 验证，采用轨迹级和样本级评估器，结合视觉关键点和多模型投票，以确保训练时强化学习（RL）信号的可靠性。

Furthermore, we integrate safety-aware mechanisms to ensure controlled execution of consequential actions. By offering a capable, efficient, and open-source foundation, UI-Venus-2 advances the field toward more generalizable, verifiable, and self-reflective agents for real-world applications.

此外，我们集成了安全感知机制，以确保对重要操作的受控执行。通过提供一个强大、高效且开源的基础，UI-Venus-2 推动了该领域的发展，使其向更具通用性、可验证性和自我反思能力的现实应用智能体迈进。

Paper Details:

Authors: Venus Team, et al.
Submission Date: 27 Aug 2026
Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
DOI: 10.48550/arXiv.2609.00028

论文详情：

作者： Venus 团队等
提交日期： 2026 年 8 月 27 日
学科分类： 人工智能 (cs.AI)；计算与语言 (cs.CL)；计算机视觉与模式识别 (cs.CV)；机器学习 (cs.LG)
DOI: 10.48550/arXiv.2609.00028