OpenMobile-2 is a new mobile agent framework that aims for near-frontier performance while keeping its training environments and recipes fully open. The project is designed to make it easier for researchers to reproduce and build upon state-of-the-art mobile agent systems.
The first of its three key advances is a diverse set of environments built on simulated commercial apps, collected in a platform called MobileGym++. The abstract also highlights app-native tools as a core component, though the details of the remaining advances are cut off in the available abstract.
By releasing both the environments and the training recipes, OpenMobile-2 addresses a common gap in mobile agent research, where many systems rely on proprietary or closed setups. This openness could accelerate progress in the field by providing a shared, transparent foundation for experimentation.