Alpamayo, a new suite of open-source AI models, simulation tools, and datasets for training real robots and cars that are intended to assist autonomous vehicles in making sense of challenging driving scenarios, was introduced by Nvidia at CES 2026.
According to a statement from Nvidia CEO Jensen Huang, “the ChatGPT moment for physical AI is here – when machines begin to understand, reason, and act in the real world.” “Alpamayo gives autonomous cars the ability to reason, which enables them to anticipate uncommon situations, drive safely in challenging situations, and provide justification for their actions.”
At the heart of Nvidia’s new family is the Alpamayo 1, a reason-based vision language action (VLA) model with 10 billion parameters that enables an autonomous vehicle (AV) to think more like a human and solve complex edge cases, such as navigating a traffic light outage at a busy intersection, without prior knowledge.
During a news conference on Monday, Ali Kani, vice president of automotive at Nvidia, stated, “It does this by breaking down problems into steps, reasoning through every possibility, and then selecting the safest path.”
In Huang’s words during his Monday keynote address, “Alpamayo not only takes sensor input and activates steering wheel, brakes, and acceleration, it also reasons about what action it’s about to take.” It informs you of the course of action it intends to pursue and the factors that led to it. The trajectory comes next, of course.
Hugging Face has the underlying code for Alpamayo 1. Developers can use Alpamayo to train simpler driving systems, refine it into smaller, faster versions for vehicle development, or build tools on top of it, such as evaluators that determine whether a car made a wise choice or auto-labeling systems that automatically tag video data.
According to Kani, “they can also use Cosmos to generate synthetic data and then train and test their Alpamayo-based AV application on the combination of the real and synthetic dataset.” Nvidia’s generative world models, or Cosmos, are AI systems that simulate the real world so they can anticipate events and respond accordingly.
As part of the Alpamayo deployment, Nvidia is also making available an open dataset that includes over 1,700 hours of driving data gathered in a variety of locations and circumstances, encompassing uncommon and intricate real-world situations. AlpaSim, an open-source simulation framework for autonomous driving system validation, is also being introduced by the company. Developers can safely test systems at scale with AlpaSim, which is available on GitHub and is made to replicate real-world driving situations, including traffic and sensors.