Hugging Face has published a research paper titled MintAct: A Unified Visual Agent for Digital Environments, introducing a family of vision-language models designed to act inside digital environments rather than specialize in a single task.
According to the paper, MintAct unifies three capabilities that are usually handled by separate systems: UI grounding, multi-step navigation across mobile, desktop, and web, and visual tool use. The models are trained at 2B, 4B, and 8B scales, and the authors state that careful design of environments, data, and training recipes allows MintAct to match the performance of per-domain specialists across all of these capabilities.
To support this, the team describes a scalable environment and reinforcement learning (RL) infrastructure. On the environment side, hundreds of concurrent instances are hosted across heterogeneous per-domain backends, serving both trajectory data collection and online RL. For training, an asynchronous framework is said to keep explicit control over the cross-domain training distribution while remaining stable under noisy environment feedback and off-policy drift.
The paper reports that MintAct achieves state-of-the-art performance across a wide range of benchmarks at comparable model sizes, citing a score of 48.9 on OSWorld-Verified.