NVIDIA used NeurIPS 2025 to unveil a major expansion of its open AI ecosystem for both digital and physical AI. The company announced new open-source models, datasets, and frameworks spanning autonomous driving, robotics, speech, and AI safety. Official post: NVIDIA Blog.

• Alpamayo-R1 for physical AI: NVIDIA introduced DRIVE Alpamayo-R1 (AR1), billed as the world’s first open, industry-scale reasoning vision-language-action (VLA) model for autonomous driving. Built on Cosmos Reason, AR1 combines chain-of-thought reasoning with path planning to handle complex road scenarios and target level 4 autonomy. :contentReference[oaicite:0]{index=0}
• Open tooling for AV research: AR1 will be released on GitHub and Hugging Face, alongside open datasets in the NVIDIA Physical AI Open Datasets collection and the new AlpaSim framework for evaluation and simulation. :contentReference[oaicite:1]{index=1}
• Cosmos Cookbook & ecosystem: NVIDIA launched the Cosmos Cookbook with recipes for using Cosmos-based models across physical AI tasks, plus open projects like LidarGen, Omniverse NuRec Fixer, Cosmos Policy, and ProtoMotions3 for lidar generation, simulation cleanup, robot policies, and digital humans. :contentReference[oaicite:2]{index=2}
• Nemotron for digital AI: On the digital side, NVIDIA added multi-speaker speech models (MultiTalker Parakeet, Sortformer), a Nemotron Content Safety Reasoning model, new synthetic safety datasets, and open-source libraries like NeMo Gym and NeMo Data Designer for RL environments and synthetic data generation. :contentReference[oaicite:3]{index=3}
• Recognition for openness: Independent group Artificial Analysis ranked the NVIDIA Nemotron family among the most open frontier AI technologies, based on licensing, data transparency, and technical documentation. :contentReference[oaicite:4]{index=4}

Why it matters: As concerns grow about closed frontier models, NVIDIA is betting on an open model stack that spans robots, self-driving cars, speech systems, and safety tooling. By releasing industry-scale physical AI models like AR1 and open digital AI components under permissive terms, NVIDIA is positioning itself as a backbone provider for researchers and startups building the next wave of agentic, real-world AI systems.