NVIDIA announced a broad release of open models, datasets and development tools aimed at accelerating Artificial Intelligence research and deployment across language, robotics and biology. The contributions span the Nemotron family for reasoning, the Cosmos platform for physical Artificial Intelligence, Isaac GR00T for robotics and Clara for biomedical AI. NVIDIA said it is sharing models and data via Hugging Face and other platforms to expand access and support developer adoption.
The Nemotron releases focus on efficient reasoning for specialized agents and include Nemotron Nano 3, Nemotron Nano 2 VL for multimodal document and video analysis, Nemotron Parse for extracting text and tables, Nemotron Safety Guard for multilingual moderation and Nemotron retrieval-augmented generation models with unified retrieval across text, images, audio and video. NVIDIA also published Nemotron training datasets (multimodal, multilingual personas and privacy-preserving synthetic personal information) and new NeMo tools such as NeMo Data Designer for synthetic data and NeMo-RL for post-training and reinforcement learning. Leading software companies are already adopting Nemotron-based agentic systems, including ServiceNow with Apriel 2.0, Palantir, Cadence, CrowdStrike, PayPal, Synopsys and Zoom.
For physical Artificial Intelligence and robotics, NVIDIA updated Cosmos and Isaac GR00T models. Releases include Cosmos Predict 2.5 for rapid world simulation, Cosmos Transfer 2.5 for higher-fidelity photorealistic data, Cosmos Reason as a vision language microservice and Cosmos Dataset Search to speed scenario retrieval. Isaac GR00T N1.6 advances humanoid reasoning and control. NVIDIA also released a large open physical AI dataset with 1,700 hours of multimodal driving sensor data and top-ranked GR00T training data used to generate synthetic data and train robots.
In healthcare and life sciences, new Clara models include CodonFM for RNA sequence modeling, La-Proteina for 3D protein structure generation and Clara Reason for chain-of-thought radiology reasoning. Select Nemotron and Cosmos models trained on DGX Cloud are available on build.nvidia.com, Hugging Face, OpenRouter and Microsoft Azure AI Foundry, with additional cloud integrations planned. Models are offered as NVIDIA NIM microservices for deployment on DGX Cloud or any NVIDIA-accelerated infrastructure to support privacy and scalable production use.
