Nvidia Nemotron 3 Ultra Prioritizes Speed and Openness for Agentic AI

Nvidia has launched Nemotron 3 Ultra, a large language model built on a hybrid transformer-Mamba architecture, designed for long-running agentic tasks. While not top-tier in overall performance compared to models like DeepSeek V4 Pro or GLM-5.2, it is notably faster than comparable open-weights models and ranks highest among U.S. open-weights models in intelligence. Nvidia has open-sourced its weights, training data, and reinforcement learning environments.