Surfer 2: The Next Generation of Cross-Platform Computer Use Agents
Abstract
Building agents that generalize across web, desktop, and mobile environments remains an open challenge, as prior systems rely on environment-specific interfaces that limit cross-platform deployment. We introduce Surfer 2, a unified architecture operating purely from visual observations that achieves state-of-the-art performance across all three environments. Surfer 2 integrates hierarchical context management, decoupled planning and execution, and self-verification with adaptive recovery, enabling reliable operation over long task horizons. Our system achieves 97.1% accuracy on WebVoyager, 69.6% on WebArena, 60.1% on OSWorld, and 87.1% on AndroidWorld, outperforming all prior systems without task-specific fine-tuning. With multiple attempts, Surfer 2 exceeds human performance on all benchmarks. These results demonstrate that systematic orchestration amplifies foundation model capabilities and enables general-purpose computer control through visual interaction alone, while calling for a next-generation vision language model to achieve Pareto-optimal cost-efficiency.
Cite
@article{arxiv.2510.19949,
title = {Surfer 2: The Next Generation of Cross-Platform Computer Use Agents},
author = {Mathieu Andreux and Märt Bakler and Yanael Barbier and Hamza Benchekroun and Emilien Biré and Antoine Bonnet and Riaz Bordie and Nathan Bout and Matthias Brunel and Aleix Cambray and Pierre-Louis Cedoz and Antoine Chassang and Gautier Cloix and Ethan Connelly and Alexandra Constantinou and Ramzi De Coster and Hubert de la Jonquiere and Aurélien Delfosse and Maxime Delpit and Alexis Deprez and Augustin Derupti and Mathieu Diaz and Shannon D'Souza and Julie Dujardin and Abai Edmund and Michael Eickenberg and Armand Fatalot and Wissem Felissi and Isaac Herring and Xavier Koegler and Erwan Le Jumeau de Kergaradec and Aurélien Lac and Maxime Langevin and Corentin Lauverjat and Antonio Loison and Avshalom Manevich and Axel Moyal and Axel Nguyen Kerbel and Marinela Parovic and Julien Revelle and Guillaume Richard and Mats Richter and Ronan Riochet and María Santos and Romain Savidan and Laurent Sifre and Maxime Theillard and Marc Thibault and Ivan Valentini and Tony Wu and Laura Yie and Kai Yuan and Jevgenij Zubovskij},
journal= {arXiv preprint arXiv:2510.19949},
year = {2025}
}
Comments
21 pages, 9 figures, 2 tables