OpenMobile-2 proves open-source smartphone agents can rival proprietary models

Researchers have developed OpenMobile-2, an open-source smartphone agent trained across an environment of 110 applications. By focusing on cross-app training rather than raw data volume, the model demonstrates performance levels comparable to proprietary alternatives on Android devices. This development highlights a shift in mobile AI training methodologies, suggesting that high-functioning autonomous agents can be built using open-source frameworks rather than exclusively relying on closed, corporate systems.