NVIDIA Nemotron 3 Super Ships: 120B-Parameter Open-Weight Agentic Model with 10T Training Tokens Under Permissive License

NVIDIA Newsroom·medium signal

NVIDIA released Nemotron 3 Super, a 120-billion-parameter open-weight model designed for complex multi-agent systems, with 10 trillion tokens of training data published alongside. The model uses a hybrid latent mixture-of-experts (MoE) architecture optimized for agentic AI. Early adopters include Cursor, CrowdStrike, Deloitte, Perplexity, and Siemens. Available under a permissive license for workstation, data center, or cloud deployment.