Training Open Models for Agentic Phone Use

The article presents PhoneBuddy, a novel training approach for developing open models capable of executing tasks on mobile phones. It integrates a real-app environment with a mock-app environment, PhoneWorld, to enhance training efficiency. Evaluation results indicate a significant improvement in task success rates, increasing from 36.67% to 45.33% on real phones and from 60.3% to 83.2% on the mock environment, highlighting the importance of combining real-app reinforcement learning with mock-app training to optimize agentic phone use.