EvoCUA: Evolving Computer Use Agents via Learning from Scalable Synthetic Experience
Taofeng Xue , Chong Peng , Mianqiu Huang , Linsen Guo , Tiancheng Han , Haozhe Wang , Jianing Wang , Xiaocheng Zhang , Xin Yang , Dengchang Zhao , Jinrui Ding , Xiandi Ma , Yuchen Xie , Peng Pei , Xunliang Cai , Xipeng Qiu
- 🏛 Institutions
- Meituan , Fudan , Tongji University , HKUST
- 📅 Date
- January 22, 2026
- 📑 Publisher
- arXiv
- 💻 Env
- Desktop
- 🔑 Keywords
TLDR
EvoCUA replaces static imitation with an evolving training loop built on verifiable task synthesis, high-throughput sandbox rollouts, and iterative policy optimization from both successful and failed trajectories. On OSWorld it reaches 56.7% success, outperforming prior open-source computer-use agents and even some leading closed-weight systems.
Related papers (24)
- IntentScore: Intent-Conditioned Action Evaluation for Computer-Use AgentsApril 6, 2026 · arXiv
- ComputerRL: Scaling End-to-End Online Reinforcement Learning for Computer Use AgentsAugust 19, 2025 · ICLR 2026 (Poster)
- GUIDE: Resolving Domain Bias in GUI Agents through Real-Time Web Video Retrieval and Plug-and-Play AnnotationMarch 27, 2026 · arXiv
- Mobile-Agent-v3.5: Multi-platform Fundamental GUI AgentsFebruary 15, 2026 · arXiv
- OmegaUse: Building a General-Purpose GUI Agent for Autonomous Task ExecutionJanuary 28, 2026 · arXiv
- CaMeLs Can Use Computers Too: System-level Security for Computer Use AgentsJanuary 14, 2026 · arXiv
- ShowUI-π: Flow-based Generative Models as GUI Dexterous HandsDecember 31, 2025 · arXiv
- MAI-UI Technical Report: Real-World Centric Foundation GUI AgentsDecember 26, 2025 · arXiv
- Step-GUI Technical ReportDecember 17, 2025 · arXiv
- UltraCUA: A Foundation Model for Computer Use Agents with Hybrid ActionOctober 20, 2025 · arXiv
- Watch and Learn: Learning to Use Computers from Online VideosOctober 6, 2025 · CVPR 2026
- Scaling Agents for Computer UseOctober 2, 2025 · arXiv
- Efficient Multi-turn RL for GUI Agents via Decoupled Training and Adaptive Data CurationSeptember 28, 2025 · arXiv
- ScaleCUA: Scaling Open-Source Computer Use Agents with Cross-Platform DataSeptember 18, 2025 · ICLR 2026 (Oral)
- Mobile-Agent-v3: Fundamental Agents for GUI AutomationAugust 21, 2025 · arXiv
- UI-Venus Technical Report: Building High-performance UI Agents with RFTAugust 14, 2025 · arXiv
- OpenCUA: Open Foundations for Computer-Use AgentsAugust 12, 2025 · NeurIPS 2025 (Spotlight)
- Evolving in Tasks: Empowering the Multi-modality Large Language Model as the Computer Use AgentAugust 6, 2025 · arXiv
- CoAct-1: Computer-using Multi-Agent System with Coding ActionsAugust 5, 2025 · ICLR 2026 (Poster)
- DPO Learning with LLMs-Judge Signal for Computer Use AgentsJune 3, 2025 · arXiv
- LiteCUA: Computer as MCP Server for Computer-Use Agent on AIOSMay 24, 2025 · arXiv
- Efficient Agent Training for Computer UseMay 20, 2025 · ICLR 2026 (Poster)
- STEVE: A Step Verification Pipeline for Computer-use Agent TrainingMarch 16, 2025 · arXiv
- SpiritSight Agent: Advanced GUI Agent with One LookMarch 5, 2025 · CVPR 2025 (Poster)