American tech giant NVIDIA launched Alpamayo 2 Super, its frontier open reasoning vision-language-action (VLA) model for autonomous vehicles (AVs) on Tuesday.
NVIDIA claimed the open reasoning VLM model helps AVs do more than just object detection and motion prediction by letting the vehicle “understand the situation, reason about cause and effect, choose the right action and turn decisions into a safe, comfortable path.”
“Beyond seeing, Alpamayo understands and reasons through the complex world – thinks before it acts. It’s a powerful backbone for robotaxis, trucks, shuttles, delivery vans, tractors and the long tail of mobile robots—billions of autonomous machines someday. We’re releasing it for commercial use under OpenMDW-1.1 so teams can inspect it, fine-tune it and deploy it—open models advance safety and security,” said Jensen Huang in a post on X.
Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles.
Beyond seeing, Alpamayo understands and reasons through the complex world – thinks before it acts.
It’s a powerful backbone for robotaxis, trucks, shuttles, delivery vans,… pic.twitter.com/K6jS0XDR4b
— Jensen Huang (@JensenHuang) August 4, 2026
In a blogpost, NVIDIA Alpamayo 2 Super model, built on NVIDIA Cosmos 3 Super Reasoner, and released on Hugging Face, supports a wide range of autonomous vehicles (AV) relevant capabilities. NVIDIA’s Cosmos 3 Super Reasoner is a 32 billion parameter autoregressive vision-language model (VLM).
NVIDIA claimed the Alpamayo 2 Super Model has been post-trained with reinforcement learning to advance the AV ecosystem on two fronts; open commercial licensing and leading multi-task capabilities for autonomous driving.
NVIDIA claimed the model targets applications like robotaxis, trucks, shuttles, and other mobile robots, with an open commercial release under OpenMDW-1.1 to enable inspection, fine-tuning, and deployment for improved safety.
NVIDIA said Alpamayo Super 2 weights are available on Hugging Face and inference code on GitHub under the permissive OpenMDW-1.1 license, which allows commercial inspection, fine-tuning, derivative works, redistribution, and deployment with no field-of-use restrictions. NVIDIA also claimed that its VLM ranked first on LingoQA (79.2 score) among nearly 40 models and leads multiple AV benchmarks in meta-action IoU, auto-labeling accuracy, visual grounding, and closed-loop AlpaSim evaluations.
Also Read: Explained: How NVIDIA Became the Company Every AI Model Depends On
