NVIDIA has released Alpamayo 2 Super, an open 32B vision-language-action model for autonomous driving, and says the model is available now for commercial use.
The August 4 launch moves Alpamayo from a research signal into a more practical development path for robotaxi and autonomous-vehicle teams. NVIDIA says the model is released under its OpenMDW-1.1 license, can be fine-tuned, and can be used to build derivative models for commercial redistribution.
The company is positioning Alpamayo 2 Super as part of a larger “cloud-to-car” stack rather than a standalone model file. The model is meant to reason over driving scenes, explain decisions, and connect with NVIDIA’s simulation tools before a system reaches road testing.
The commercial license is the shift
Open autonomous-driving models are useful only if developers can move beyond demo evaluation. NVIDIA’s launch says Alpamayo 2 Super is available on Hugging Face and can be used commercially, which changes the procurement question for teams already building around the DRIVE ecosystem.
That does not make the model a finished driver. It makes it a testable component in a regulated physical-world workflow. Developers still have to validate perception, planning, controls, fallback behavior, safety cases, and domain transfer from simulation to real roads.
NVIDIA says the broader Alpamayo model family has passed 500,000 downloads on Hugging Face. That is a usage signal, not proof of deployment. It does suggest that developers are already treating the family as a reference point for open autonomous-driving research.
Simulation is part of the product story
The launch connects Alpamayo 2 Super to NVIDIA’s autonomous-vehicle simulation stack. NVIDIA’s developer page describes Omniverse NuRec for reconstructing real-world scenes, Cosmos-Dreams for generating scenario variations, and AlpaSim for closed-loop testing across virtual scenarios.
That matters because autonomous driving is not only a model-quality problem. A driving model needs to be tested against rare events, weather shifts, lighting changes, sensor artifacts, road geometry, traffic behavior, and policy decisions that are hard to collect safely at scale.
The useful read is that NVIDIA is selling an inspectable model plus the environment around it: training data workflows, simulation, synthetic scenarios, validation, and in-vehicle deployment hardware.
