NVIDIA unveils Alpamayo family of open models and simulation tools

NVIDIA unveiled the Alpamayo family, an open collection of Artificial Intelligence (AI) models, simulation tools and datasets intended to speed development of reasoning-based level 4 autonomous vehicle capabilities.

The company said autonomous vehicles must operate across a wide range of driving conditions and that rare, complex “long-tail” scenarios remained difficult to address. NVIDIA described limits in architectures that separated perception and planning and noted recent end-to-end learning advances had not fully resolved edge-case reasoning about cause and effect.

Alpamayo combined Chain of Thought (CoT) reasoning-based vision language action models, simulation frameworks and open datasets and referenced the NVIDIA Halos safety system as underpinning those components. NVIDIA positioned the models as large-scale teacher models for fine-tuning and distillation, and at CES released Alpamayo 1 (a 10-billion-parameter model on Hugging Face with video-based trajectories, reasoning traces, open weights and inferencing scripts), AlpaSim (an open-source simulation framework on GitHub) and the Physical AI Open Datasets (1,700+ hours of driving data on Hugging Face).

Mobility organizations including Lucid, JLR, Uber and Berkeley DeepDrive showed interest in using the releases to develop reasoning-based level 4 stacks. NVIDIA said developers could access Test Access Points (TAP) into its tool and model library from NVIDIA Cosmos and NVIDIA Omniverse, fine-tune models on proprietary fleet data, integrate them with NVIDIA DRIVE Hyperion and NVIDIA DRIVE AGX Thor accelerated compute, and validate performance in simulation before commercial deployment.

“The ChatGPT moment for physical AI is here — when machines begin to understand, reason and act in the real world,” said Jensen Huang, founder and CEO of NVIDIA. “Robotaxis are among the first to benefit. Alpamayo brings reasoning to autonomous vehicles, allowing them to think through rare scenarios, drive safely in complex environments and explain their driving decisions — it’s the foundation for safe, scalable autonomy.”

The press release included forward-looking statements, listed factors that could cause actual results to differ and said NVIDIA disclaimed any obligation to update those statements.