NVIDIA releases new open-source AI models for speech, safety, and autonomous driving at NeurIPS 2025.
At NeurIPS 2025, NVIDIA unveiled a suite of open-source AI models and tools, including Alpamayo-R1, the world’s first open industry-scale reasoning vision language action (VLA) model for autonomous driving. The company also introduced new models for speech recognition, AI safety, and synthetic dataset generation, expanding its digital and physical AI developer toolkit.
Among the new releases are LidarGen for AV simulation, Omniverse NuRec Fixer for robotics, and NeMo Gym for reinforcement learning environments. These tools are designed to accelerate research and development in autonomous systems, robotics, and generative AI. By open-sourcing these resources, NVIDIA aims to empower researchers and developers worldwide to build more advanced and reliable AI applications.
Source: NVIDIA Blog