Nvidia unveils family of open AI models for AVs, robots

Nvidia used CES 2026 to launch Alpamayo, a family of open AI models, tools and datasets aimed at advancing autonomous vehicles and robots. CEO Jensen Huang described Alpamayo as a "thinking, reasoning" AV AI, trained end-to-end from camera input to vehicle control, designed to handle complex and rare traffic situations rather than just repeat scenarios it has already seen.

The core model, Alpamayo 1, is a 10‑billion‑parameter vision-language-action system that Nvidia positions as a large "teacher" model. Instead of running directly in cars, developers can fine-tune it into smaller runtime models and use it to build tools such as reasoning-based evaluators and auto-labelling systems. Nvidia is publishing Alpamayo’s code and datasets on Hugging Face, including more than 1,700 hours of driving data covering varied conditions and edge cases, and releasing AlphaSim, an open-source simulation framework on GitHub with realistic sensor modelling and configurable traffic for closed-loop AV testing. Huang said he expects a large share of cars to be autonomous within a decade and argued that the same synthetic data and simulation methods will apply across many robotic systems.