At CES 2026 this year, NVIDIA once again showcased its technological prowess in the field of autonomous driving, announcing the launch of a series of open-source AI models and simulation tools called "Alpamayo," which will solve the most challenging "long tail" problem in autonomous vehicle development. Furthermore, NVIDIA confirmed that its full-platform autonomous driving software, DRIVE AV, will be officially installed in the new Mercedes-Benz CLA model later this year, and will further expand the DRIVE Hyperion ecosystem, collaborating with multiple leading suppliers to accelerate the deployment of Level 4 autonomous driving technology.
"Alpamayo": Enabling self-driving cars to "think like humans".
The core purpose of NVIDIA's newly launched "Alpamayo" open-source model family is to address the challenges faced by autonomous driving systems in rare or complex scenarios (i.e., the long tail effect). By endowing vehicles with the ability to perceive, infer, and make human-like judgments, it significantly improves safety.
Alpamayo mainly consists of three key components:
• Alpamayo 1 model: This is the industry's first open-source inferential visual language action (VLA) model. It has a parameter architecture of 100 billion sets and can generate driving paths based on image input, while simultaneously producing "reasoning traces" that explain the logic behind the driving decision.
• AlpaSim Simulator: A fully open-source, high-fidelity simulation framework that supports realistic sensor modeling and configurable traffic dynamics, allowing developers to quickly validate their models.
• Physical AI Open Dataset: Contains over 1700 hours of large-scale driving data, covering diverse geographical environments and road conditions, used to train the inference capabilities of models.
Currently, companies including Jaguar Land Rover, Lucid, Uber, and academic research institutions such as Berkeley DeepDrive have planned to use Alpamayo to advance their Level 4 autonomous vehicle development blueprints.
DRIVE AV software debuts: Mercedes-Benz's new CLA is the first to feature it.
Regarding its application in mass-produced vehicles, NVIDIA announced that its DRIVE AV software will officially debut on US roads this year, with the first vehicle being the all-new Mercedes-Benz CLA.
This software will provide enhanced Level 2 peer-to-peer driving assistance features, including urban navigation in complex environments, active collision avoidance, and automatic parking. The system employs a "dual-stack" architecture, consisting of an end-to-end AI stack responsible for core driving functions, paired with a traditional safety stack based on the NVIDIA Halos safety system. This parallel operation ensures system redundancy and safety.
DRIVE Hyperion ecosystem expansion: Aiming for Level 4 autonomous driving
To address the demands of more advanced autonomous driving, NVIDIA announced the expansion of its DRIVE Hyperion global ecosystem. New partners include leading sensor manufacturers such as Sony, Bosch, Hesai, Magna, and ZF Group.
The DRIVE Hyperion platform is powered by two Blackwell-based DRIVE AGX Thor system-on-a-chip (SoCs), delivering over 2000 TFLOPS (FP4) of real-time computing performance. All deployments are based on NVIDIA Halos, a comprehensive security and cybersecurity framework that ensures the autonomous driving system undergoes rigorous verification and certification.
