NVIDIA unveiled five open-source AI model families at GTC 2026, marking an expansion from language processing into robotics, autonomous systems, and scientific research.1
The release includes Cosmos 3, a world foundation model for simulation; Isaac GR00T N1.7 and N2 for robotics applications; Nemotron 3 language models; Alpamayo 1.5; and Proteina-Complexa for protein design.2 The models target developers building intelligent agents across digital and physical industries.
"Open source AI has become a global force for innovation," said Kari Briski, NVIDIA's vice president of product management.3 "From biology and scientific discovery to robotics and autonomous machines, NVIDIA open model families extend intelligence beyond language, enabling developers worldwide to build intelligent agents and power breakthroughs across digital and physical industries."3
The Isaac GR00T models address physical AI applications, allowing robots to learn from visual demonstrations and adapt to real-world environments. Cosmos 3 provides world simulation capabilities for training autonomous systems without requiring extensive real-world data collection.
Proteina-Complexa targets drug discovery and biological research, generating protein structures for therapeutic development. The model joins NVIDIA's scientific AI portfolio alongside Alpamayo 1.5.
NVIDIA partnered with Adobe, LangChain, and CrowdStrike to accelerate enterprise adoption.4 The partnerships aim to integrate the models into existing development workflows and cybersecurity systems.
The release shifts NVIDIA's open-source strategy beyond its established position in language models. Physical AI and scientific discovery models address compute-intensive applications where NVIDIA's GPU infrastructure provides advantages over CPU-based alternatives.
Developers can access the models through NVIDIA's AI platform, with pre-trained weights and training frameworks available for customization. The company provides reference implementations for common use cases in robotics, autonomous vehicles, and laboratory automation.