NVIDIA Launches Alpamayo to Power Reasoning-Based Autonomous Driving

NVIDIA unveiled the Alpamayo family of open AI models and simulation tools at CES today, signaling a major shift toward reasoning-based autonomous vehicle development. This suite aims to solve the toughest challenges in self-driving, specifically the rare and complex scenarios known as the long tail. By providing open-source tools, NVIDIA hopes to enable vehicles that can perceive, reason, and act with judgment similar to a human driver. The core of the release is Alpamayo 1, a 10-billion-parameter model designed for the research community. This vision language action model processes video to generate driving paths while explaining its logic step by step. Developers can find the model weights and inferencing scripts on Hugging Face. Alongside the model, NVIDIA is releasing AlpaSim and a massive set of Physical AI Open Datasets. AlpaSim is a fully open-source simulation framework available on GitHub for high-fidelity testing. The datasets include over 1,700 hours of driving data covering diverse geographies and rare real-world edge cases available on Hugging Face.

The Alpamayo family includes chain-of-thought reasoning models that bring humanlike thinking to vehicle decision-making. These systems are underpinned by the NVIDIA Halos safety system to ensure judgment remains within safe bounds. Jensen Huang, founder and CEO of NVIDIA, described the launch as the ChatGPT moment for physical AI where machines begin to understand and reason in the real world.

Major industry players including Lucid, JLR, and Uber have already expressed support for the Alpamayo ecosystem. These companies view reasoning models as critical for achieving safe Level 4 autonomous deployment. The open nature of the project also empowers academic researchers at institutions like Berkeley DeepDrive to innovate at an unprecedented scale.