At CES 2026, Nvidia unveiled Alpamayo, a groundbreaking family of open-source AI models, simulation tools, and datasets. This new suite is designed to equip physical robots and autonomous vehicles with advanced reasoning capabilities, enabling them to navigate complex driving situations with human-like intelligence.
“The ChatGPT moment for physical AI is here — when machines begin to understand, reason, and act in the real world,” stated Nvidia CEO Jensen Huang. “Alpamayo brings reasoning to autonomous vehicles, allowing them to think through rare scenarios, drive safely in complex environments, and explain their driving decisions.”
Central to this new family is Alpamayo 1, a 10-billion-parameter, chain-of-thought, Vision Language Action (VLA) model. This model empowers autonomous vehicles (AVs) to process information and make decisions more akin to human drivers, tackling intricate "edge cases" without prior experience. For instance, an AV powered by Alpamayo 1 could logically navigate a traffic light outage at a busy intersection.
Ali Kani, Nvidia’s vice president of automotive, explained during a press briefing, “It does this by breaking down problems into steps, reasoning through every possibility, and then selecting the safest path.” Huang further elaborated during his keynote, highlighting Alpamayo’s ability not only to interpret sensor input and control vehicle mechanics but also to reason about its intended actions, explain its decisions, and predict the resulting trajectory.
Empowering Developers with Open-Source Tools
Nvidia is making Alpamayo 1’s underlying code publicly available on Hugging Face. This open-source approach allows developers to fine-tune Alpamayo into smaller, faster versions optimized for specific vehicle development needs. They can also use it to train simpler driving systems or build advanced tools, such as auto-labeling systems for video data or evaluators that assess the intelligence of a car's decisions.
Furthermore, developers can leverage Nvidia’s generative world model, Cosmos, to generate synthetic data. This synthetic data can then be combined with real-world datasets to train and rigorously test Alpamayo-based AV applications, enhancing their robustness and reliability.
Comprehensive Ecosystem for Autonomous Driving
Alongside Alpamayo, Nvidia is releasing a comprehensive open dataset featuring over 1,700 hours of diverse driving data. Collected across various geographies and conditions, this dataset includes rare and complex real-world scenarios crucial for advanced AV training.
The company is also launching AlpaSim, an open-source simulation framework available on GitHub. AlpaSim is designed to accurately recreate real-world driving conditions, from sensor inputs to traffic dynamics, enabling developers to safely and efficiently validate autonomous driving systems at scale.
