NVIDIA

Autonomous Vehicles · AI Models · Robotics

Launch

NVIDIA opens Alpamayo 2 Super robotaxi model for commercial use

Giving the model away for free deepens reliance on the specialized chips needed to run it.

• NVIDIA released Alpamayo 2 Super for commercial use on August 4, 2026, lifting the licensing restriction that had kept its Alpamayo autonomous-driving model family in research-only status.
• The model is a roughly 34-billion-parameter vision-language-action system pairing NVIDIA's Cosmos 3 Super Reasoner with a diffusion-based action expert to turn camera video into driving decisions.
• It processes 360-degree video from up to seven cameras and outputs trajectories, 'chain-of-causation' reasoning traces, meta-actions like yield or lane-change, and grounded scene answers.
• Weights are posted on Hugging Face under OpenMDW-1.1, the Linux Foundation's permissive license, covering fine-tuning, derivative models and commercial redistribution.
• Alpamayo 2 Super is a cloud teacher model, not in-car software; developers distill its outputs into smaller models tuned to run on NVIDIA's DRIVE Thor and Hyperion in-vehicle chips.
• NVIDIA says Alpamayo models have been downloaded roughly 400,000 to 500,000 times since the family launched in January, and the stack already underpins Uber's robotaxi effort in Munich.
• By open-sourcing the reasoning layer, NVIDIA pushes further into the AV software stack while keeping its real leverage in the compute those models are trained and deployed on.