NVIDIA is expanding its offerings in open-source AI technologies, aiming to support research in both digital and physical AI domains. At the NeurIPS conference, they introduced Alpamayo-R1, the first open reasoning vision language action (VLA) model tailored for autonomous vehicle (AV) research. This model integrates chain-of-thought reasoning with path planning, enhancing AV decision-making in complex scenarios, such as navigating busy intersections. Researchers can customize AR1 for non-commercial projects, and it’s already available on GitHub and Hugging Face, with some training data accessible through NVIDIA's open datasets. The article highlights several tools and models that NVIDIA has released to bolster its AI ecosystem. In the physical AI space, the Cosmos Cookbook provides developers with resources to use and post-train Cosmos-based models. Notable applications include LidarGen, which generates lidar data for AV simulations, and Cosmos Policy, a framework for developing robot behavior models. In digital AI, NVIDIA is rolling out new speech AI models and tools for generating synthetic datasets, like MultiTalker Parakeet, which can recognize multiple speakers in real-time, and Nemotron Content Safety, which enforces safety policies in audio content. NVIDIA's commitment to open-source development is reinforced by its recognition in a new Openness Index, which benchmarks AI technologies for their transparency and accessibility. At NeurIPS, the company is showcasing over 70 research papers and projects, with a focus on advancements in AI reasoning and language models. Innovations like Audio Flamingo 3, which processes lengthy audio segments, and Minitron-SSM, a method for compressing language models, highlight NVIDIA's ongoing contributions to the AI field. This emphasis on openness and collaboration positions NVIDIA as a key player in the evolving AI research landscape.