Researchers introduce Surfer 2, a unified architecture for computer use agents that operates purely from visual observations across web, desktop, and mobile environments. The system integrates hierarchical context management, decoupled planning and execution, and self-verification with adaptive recovery to enable reliable operation over long task horizons.
- Achieves 97.1% accuracy on WebVoyager, 69.6% on WebArena, 60.1% on OSWorld, and 87.1% on AndroidWorld without task-specific fine-tuning.
- Outperforms all prior systems and exceeds human performance on all benchmarks with multiple attempts.
The results demonstrate that systematic orchestration amplifies foundation model capabilities to enable general-purpose computer control through visual interaction alone.