AI News

MAI-Gym is an open platform that combines GUI-agent evaluation, data collection, and online reinforcement-learning training across mobile, computer, and web environments. It uses a shared rollout pipeline so agents behave consistently during training and evaluation, supports ten established benchmarks, and can scale to 20,000 parallel environment instances. The platform also synthesizes diverse GUI applications and difficult tasks, including resettable mocked applications and coherent cross-application worlds with execution-validated task verifiers. Using the platform, the authors train MAI-Gym-9B, a unified 9-billion-parameter agent for mobile and computer use. Among open GUI agents with fewer than 10 billion parameters, it reaches 78.4% on MobileWorld and 76.2% on OSWorld-Verified, which the paper reports as state-of-the-art results in that group. Training with reinforcement learning in a mocked train-ticketing application transfers to its real-world counterpart: task success on a physical phone rises from 28.8% to 48.1%. The authors release the framework, synthesis skills, mocked applications, 2,000 verified tasks, and the model checkpoint to support reproducible research on scalable GUI-agent training.