iOS

December 16, 2025, 6:54 AM EST. Nvidia unveils the Nemotron 3 family-Nano, Super, Ultra-open models designed for transparent, scalable agent workflows. Built on a hybrid latent MoE architecture, these models aim to boost throughput, extend context, and support multi-agent automation. Nvidia provides model weights, datasets, reinforcement learning environments and tooling to developers and enterprises, enabling on-premises usage, customization, and integration with existing systems. The move signals a broader push into open-source AI, offering greater transparency around training data and methodologies while reducing licensing friction. Open models balance capability gaps with flexibility and lower costs, appealing for bespoke edge deployments. Nvidia's openness mirrors industry trends with other players exploring open-weight releases, potentially accelerating enterprise adoption of AI agents and automation across sectors.