NVIDIA Opens Its 34-Billion-Parameter Driving Model for Commercial Use

NVIDIA has made Alpamayo 2 Super available for commercial use. It is a vision-language-action model of roughly 34 billion parameters aimed at robotaxis, trucks, shuttles and delivery vehicles. Rather than emitting only a trajectory, the model produces a reasoning trace and a plain-language account of the decision behind it, as described in coverage of the release.

The license is the operative change. Alpamayo 2 Super is distributed under OpenMDW-1.1, the Linux Foundation's permissive license for open model distributions, which covers fine-tuning, derivative models and commercial redistribution. Distilled versions can be shipped without further permission from NVIDIA, and model outputs carry no license conditions. A developer can adapt the model to fleet data and a specific driving policy, then deploy it, without negotiating separately.

The intended workflow is as much about data as about driving. NVIDIA's developer documentation presents the model as a generator of trajectories, reasoning traces and automatic labels. That addresses the expensive part of autonomy work, namely annotating rare long-tail scenarios. PlusAI has said it will use the Alpamayo family in its SuperDrive virtual driver, with factory-built driverless trucks targeted for 2027.

NVIDIA's claim that Alpamayo is the most-adopted open reasoning model family for autonomous driving on Hugging Face comes from the company itself. No independent benchmark of Alpamayo 2 Super against production stacks from Waymo, Tesla or Chinese suppliers accompanied the release.