Chipmaker NVIDIA has released a set of open artificial intelligence (AI) models, datasets, and developer tools to help companies build practical AI systems across sectors such as software, robotics, transportation, and healthcare.
The company said the new releases expand its open model portfolio, covering language and speech AI, robotics, autonomous vehicles, and biomedical research. By making these resources publicly available, NVIDIA aims to lower barriers for companies and researchers developing AI systems for real-world use.
Among the releases are additions to the NVIDIA Nemotron family of models, which focus on so-called agent-based AI systems that can process information, make decisions, and take actions. NVIDIA introduced new Nemotron models for speech recognition, document search, and AI safety. The speech models are designed for real-time transcription and voice-based applications, while the retrieval models help organizations search through large collections of text, images, and other data. The safety models are intended to detect harmful content and sensitive personal information in AI systems.
NVIDIA said the overall goal of these releases is to provide shared building blocks that companies and researchers can adapt to their own needs, while reducing development time and cost across industries.
NVIDIA is also releasing large open datasets and training frameworks. These include trillions of language tokens, hundreds of thousands of robotics movement records, protein structure data, and vehicle sensor data. The company said these datasets are intended to support research in areas such as robotics, science, and autonomous driving, where access to large and diverse data is often limited.
For robotics and physical AI, NVIDIA introduced the Cosmos family of open models. These models are designed to help machines understand and operate in physical environments. NVIDIA said the Cosmos models support tasks such as visual understanding, reasoning, and the creation of simulated environments for training. Companies including Salesforce, Hitachi, and Uber are using Cosmos models for applications such as traffic analysis and workplace monitoring.
NVIDIA also released updated versions of Isaac GR00T, an open model for humanoid robots. Robotics companies such as Franka Robotics and NEURA Robotics are using the model to train robots in simulation before deploying them in real-world settings.
In autonomous vehicles, NVIDIA unveiled the Alpamayo family of open models and tools. These are designed to help vehicles understand complex driving situations and explain their actions. The release includes an open simulation framework and more than 1,700 hours of driving data collected across different locations and conditions.
For healthcare and life sciences, NVIDIA introduced new Clara models focused on drug discovery and biomedical research. These models support protein design, drug safety testing, RNA structure prediction, and practical drug manufacturing considerations. NVIDIA also released a dataset of hundreds of thousands of synthetic protein structures to support further research.