Hugging Face has introduced a groundbreaking robotics model called SmolVLA, designed to bring advanced robotics capabilities to everyday developers and researchers. The new open-source AI model is so efficient that it can run on a MacBook or any device with a single consumer GPU making sophisticated robotics development more accessible than ever.
With 450 million parameters, SmolVLA is trained on Hugging Face’s LeRobot Community Datasets, which consist of community-shared and compatibly licensed data. Despite its compact size, SmolVLA reportedly outperforms larger robotics models in both simulation and real-world tests.
This launch is part of Hugging Face’s broader mission to democratize robotics. Last year, the company rolled out LeRobot, a suite of open-source tools and models for robotics. More recently, it acquired French startup Pollen Robotics and began offering low-cost robotics systems, including humanoid robots.
A unique feature of SmolVLA is its asynchronous inference stack, which separates action processing from sensory processing. This architecture allows robots to react faster in dynamic environments, making the model suitable for real-time applications.
While Hugging Face is leading with openness and affordability, the open robotics space is heating up. Companies like Nvidia, K-Scale Labs, Dyna Robotics, and Jeff Bezos-backed Physical Intelligence are also investing heavily in the future of generalist robotic agents.
With SmolVLA, Hugging Face is taking a major step toward making robotics research accessible to more developers and innovators — from labs to living rooms.
