NVIDIA has released a set of open models, datasets, and development tools covering language, agentic systems, robotics, autonomous driving, and biomedical research. The update expands several existing NVIDIA model families and makes accompanying training data and reference implementations available through GitHub, Hugging Face, and NVIDIA’s developer platforms.
In the agentic AI domain, NVIDIA extended the Nemotron model family with new components for speech recognition, retrieval-augmented generation, and safety. Nemotron Speech includes automatic speech recognition models optimized for low-latency, real-time use cases. Nemotron RAG introduces embedding and reranking vision-language models intended for multimodal document search and retrieval pipelines. Nemotron Safety adds updated models for content filtering and detection of sensitive or personally identifiable information. NVIDIA also released datasets and training code used for selected Nemotron models, including embedding models evaluated on public benchmarks.
For robotics and physical AI, NVIDIA introduced new Cosmos world foundation models, which support perception, reasoning, and synthetic data generation in real-world environments. Cosmos Reason 2 is a multimodal reasoning model designed to enhance scene understanding for agents operating in physical environments. Cosmos Transfer 2.5 and Cosmos Predict 2.5 focus on generating synthetic video data across varied environments and conditions, supporting simulation and data augmentation workflows. Built on top of Cosmos, NVIDIA released Isaac GR00T N1.6, an open vision-language-action model for humanoid robots that supports full-body control and integrates visual perception with action planning.
One component of the announcement was NVIDIA Alpamayo, a new open model family for reasoning-based autonomous driving. Alpamayo combines perception, planning, and explainability in a vision-language-action architecture and is paired with simulation tools and large-scale driving datasets. NVIDIA also introduced AlpaSim, an open-source simulation framework for closed-loop evaluation of autonomous vehicle models.
According to Xinzhou Wu, Head of Automotive at NVIDIA, Alpamayo and related tooling reflect multi-year development efforts across research, simulation, data engineering, safety, and integration teams. Wu noted that the work involved extensive road testing, continuous large-scale simulation using platforms such as Cosmos, and close collaboration with automotive partners, including Mercedes-Benz, with initial deployments planned for upcoming production vehicles.
The healthcare and life sciences updates are delivered through new NVIDIA Clara models. These include La-Proteina for atom-level protein design, ReaSyn v2 for synthesis-aware drug design, KERMT for early-stage safety and interaction prediction, and RNAPro for RNA structure modeling. NVIDIA also published a dataset of 455,000 synthetic protein structures to support training and evaluation in this domain.
All models and datasets are released under open licenses and are accessible via GitHub and Hugging Face. NVIDIA stated that many of the models are also packaged as NIM microservices for deployment on NVIDIA-accelerated systems, from local inference environments to cloud infrastructure.