Nemotron

NVIDIA Nemotron is a family of highly efficient, open-weights multimodal models engineered to power complex, long-running agentic AI workflows.

Built by NVIDIA to tackle the heavy computational demands of agentic AI, the Nemotron family (spanning Nano, Super, and Ultra tiers) delivers high-throughput reasoning and tool-use capabilities. The flagship Nemotron 3 Ultra utilizes a hybrid Mamba-Transformer Mixture-of-Experts architecture (550 billion total parameters with 55 billion active) to process up to 1 million tokens of context while maintaining native speculative decoding for fast inference. By releasing these models with open weights, open training data, and transparent recipes, NVIDIA provides developers with a highly customizable, enterprise-grade foundation for building autonomous agents that can plan, code, and execute multi-step workflows at a fraction of standard computing costs.

https://build.nvidia.com/nvidia/nemotron-4-340b-instruct