Introducing NVIDIA Alpamayo 2 Super
NVIDIA Alpamayo 2 Super is an open source 34-billion-parameter reasoning Vision-Language-Action model for Level 4 autonomous vehicle and robotaxi development. Built on the NVIDIA Cosmos 3 Super Reasoner backbone, Alpamayo 2 Super delivers 360-degree surround perception across up to seven cameras, causal reasoning in long-tail scenarios, and fleet-scale auto-labeling without manual annotation. In this video, NVIDIA Senior Director of AV Research Marco Pavone walks through the architecture, key capabilities, and how to put Alpamayo 2 Super to work in your AV development workflow. The model generates trajectory predictions alongside Chain-of-Causation reasoning traces that explain every driving decision in causal terms. It also produces Meta Actions — structured behavioral signals like yield, lane change, and stop — that downstream planners and safety tools can act on directly. Alpamayo 2 Super addresses two of the most expensive bottlenecks in AV development. First, front-camera-only models miss the full surround context required for safe merges, lane changes, and intersection navigation. Second, manual annotation of long-tail driving scenarios costs between $5 and $20 per clip, with backlogs running three to six months. With Alpamayo 2 Super, annotation cycles that used to take months now take days, and the pipeline extends to any driving domain or geography. The model is fully open and commercially available. Developers can inspect the weights, post-train on their own data, and adapt the model to their sensor configuration and operational policies. The Alpamayo family has surpassed 500,000 downloads on Hugging Face and is in use at every major OEM. Model weights are available on Hugging Face and inference notebooks are on GitHub. ➡️ Learn more: https://www.nvidia.com/en-us/solutions/autonomous-vehicles/alpamayo/ 📝 Tech blog: https://developer.nvidia.com/blog/generate-trajectories-reasoning-traces-and-auto-labels-with-nvidia-alpamayo-2-super 📥 Download: https://huggingface.co/nvidia/Alpamayo2-Super #NVIDIA #Alpamayo #AutonomousVehicles #PhysicalAI #OpenModels #AV #Robotaxi #VisionLanguageActionModel