NVIDIA Unveils Alpamayo Open AI Models for Autonomous Vehicle Reasoning

NVIDIA announced in a press release the launch of Alpamayo, a family of open-source AI models, simulation frameworks, and datasets aimed at advancing reasoning-based autonomous vehicle development. Introduced at CES 2026, Alpamayo is designed to help vehicles interpret complex environments and make decisions with humanlike judgment.

At the center of the release is Alpamayo 1, a 10-billion-parameter vision language action (VLA) model available on Hugging Face. The model uses video input to generate trajectories and reasoning traces, showing the logic behind each driving decision. Developers can fine-tune or distill the model into smaller versions for in-vehicle use or leverage it for creating tools such as auto-labeling systems and reasoning-based evaluators.

The Alpamayo ecosystem also includes AlpaSim, an open-source simulation framework available on GitHub. AlpaSim provides realistic sensor modeling and traffic dynamics for scalable testing environments. Additionally, NVIDIA released the Physical AI Open Datasets, featuring over 1,700 hours of diverse driving data from varied geographies and conditions to support training and validation of autonomous systems.

NVIDIA stated that companies including Lucid, JLR, Uber, and Berkeley DeepDrive are exploring Alpamayo to accelerate their Level 4 autonomy programs. The company noted that these open resources aim to enable developers and researchers to build safer and more explainable autonomous driving systems.