Nvidia introduces Alpamayo, an open-source AI model that allows autonomous cars to ‘think like a human’

Pa The cost of CES 2026Nvidia introduced Alpamayo, a new family of open-source AI models, simulation tools, and training datasets for physical robots and cars designed to help autonomous vehicles think through complex driving challenges.

“ChatGPT’s era of physical AI is here — when machines begin to understand, reason, and act on real-world things,” Nvidia CEO Jensen Huang said in a statement. “Alpamayo brings the concept of autonomous vehicles, allowing them to think beyond the limits, navigate in challenging environments, and explain their driving decisions.”

At the heart of Nvidia’s new family is Alpamayo 1, a 10-billion-parameter chain-of-thought model, reason-based vision language action (VLA) that allows AV to think like a human to solve complex problems – such as how to manage the flow of traffic on a busy road – without previous events.

“It does this by dividing the problem into steps, thinking through the possibilities, and choosing the best solution,” Ali Kani, Nvidia’s vice president of automotive, said Monday at a press conference.

Or as Huang said in his main article on Monday: “It (Alpamayo) doesn’t just take a sensor device and turn on the steering, brakes, and acceleration, it also thinks about what it needs to do.

The source code for Alpamayo 1 is available at Hugging Face. Developers could turn the Alpine into a small, high-speed model of a car, use it to teach simple driving techniques, or build devices on top of it such as automated video recorders or monitors that determine whether a car has made a smart choice.

“They can also use Cosmos to generate artificial data to train and test their Alpamayo-based AV application by combining real and synthetic data,” Kani said. Cosmos and Nvidia’s version of global graphicsAI systems that create a picture of the environment in order to make predictions and take action.

As part of the release of Alpamayo, Nvidia is also releasing open data with more than 1,700 hours of driving that have been collected in different environments and situations, which are complex and challenging in real-time. The company is also launching AlpaSim, an open source test system for verifying autonomous driving systems. Available on GitHub, AlpaSim is designed to recreate real-world scenarios, from sensors to cars, so developers can test systems at scale.