At CES 2026, Nvidia unveiled Alpamayo, a groundbreaking open-source ecosystem of AI models, simulation tools, and datasets designed to advance reasoning-based autonomous vehicles (AVs). The launch marks what Nvidia CEO Jensen Huang described as the “ChatGPT moment for physical AI”—a turning point where machines begin to understand, reason, and act in the real world. At the heart of the initiative is Alpamayo 1, a 10-billion-parameter vision language action (VLA) model that uses chain-of-thought reasoning to enable AVs to navigate rare and complex driving scenarios—such as a traffic light failure at a busy intersection—without prior experience. Alpamayo 1 processes video input to generate not only driving trajectories but also detailed reasoning traces that explain the decision-making process. This transparency allows developers and regulators to understand why an AV chose a particular action, enhancing trust and safety. Unlike traditional AV systems that separate perception and planning, Alpamayo integrates reasoning directly into decision-making, enabling better adaptation to edge cases that fall outside training data. The model’s code is now available on Hugging Face, with open weights and inferencing scripts, allowing developers to fine-tune it into smaller, faster versions for in-vehicle deployment. It can also serve as a foundation for building tools like auto-labeling systems for video data and evaluators that assess whether a vehicle’s decisions were safe and logical. Nvidia’s Cosmos generative world models further enhance the system by creating synthetic environments to generate training data, which can be combined with real-world data for more robust model training. Alongside Alpamayo 1, Nvidia released AlpaSim, a fully open-source simulation framework on GitHub. AlpaSim provides high-fidelity, end-to-end simulation of real-world driving conditions, including realistic sensor modeling, dynamic traffic, and closed-loop testing environments. This enables developers to safely validate AV systems at scale before real-world deployment. The ecosystem also includes a new open dataset featuring over 1,700 hours of diverse driving footage collected across varied geographies and conditions. It captures rare and complex scenarios essential for training reasoning-capable AVs. The dataset is hosted on Hugging Face, ensuring broad accessibility. The initiative has drawn strong support from industry leaders. Lucid Motors, Jaguar Land Rover (JLR), Uber, and Berkeley DeepDrive praised the open approach, emphasizing its role in accelerating innovation and enabling safer, more transparent autonomy. JLR highlighted the importance of open-source development for responsible progress, while Uber noted that Alpamayo addresses one of the biggest challenges in AVs: handling unpredictable, long-tail scenarios. Nvidia’s broader ecosystem, including the NVIDIA DRIVE Hyperion architecture powered by DRIVE AGX Thor, allows developers to integrate Alpamayo into their AV stacks, fine-tune models on proprietary fleet data, and validate performance in simulation. The company also continues to expand its tools through platforms like Omniverse and Cosmos. Alpamayo represents a major leap forward in physical AI, offering a self-reinforcing development loop that combines real and synthetic data, open models, and high-fidelity simulation. By making these tools freely available, Nvidia is empowering researchers and developers worldwide to build safer, more intelligent, and explainable autonomous systems. The launch not only advances the path to level 4 autonomy but also sets a new standard for open, collaborative innovation in the future of mobility.