Nvidia Corp. announced two new services: the Nemotron 3.5 Lightning model and an agentic AI model router named NeMo Switchyard. Nemotron 3.5 Lightning is a 30 billion-parameter mixture-of-experts model that Nvidia claims delivers up to four times the output speed and 30% faster agentic task completion compared with other models in its weight class. It is open and customizable, allowing post-training with Nvidia NeMo on enterprise hardware. Nvidia launched the Nemotron 3 model family in December with three sizes: Nano, Super and Ultra. According to Vice President of Generative AI Kari Briski, CodeRabbit Inc. used Nvidia’s standard auto model recipe, trained for one epoch, and produced a router agent for $85 in around two hours. Another partner dropped Lightning into an existing post-training stack with no changes required. Nvidia is also releasing its datasets, post-training datasets and the recipes/framework used to train the models, enabling developers to blend Nvidia’s data with their own.