NVIDIA's Nemotron 3.5 Lightning: Open-Source AI for All

NVIDIA has released Nemotron 3.5 Lightning, an open-source AI model with 30 billion parameters. This "lightweight" model is designed for agentic workflows and can run on a single GPU. What's interesting is that NVIDIA's CEO, Jensen Huang, is actively pushing for more open-weight AI models, stating they are key to America’s AI leadership. The company announced the launch in a blog post, highlighting its use for specialized tasks within larger multi-agent systems. Nemotron 3.5 Lightning is a mixture-of-experts model with 3 billion active parameters. It delivers up to four times faster output speed than similar-sized models, leading to 30% faster agentic task completion. Users can also customize it using the company's NeMo. CrowdStrike, Harvey, and CodeRabbit are among the companies that have already tested and customized this model. It's available on HuggingFace and includes weights, data, and recipes. NVIDIA also released NeMo Switchyard, an open-source library for smart routing within agent tools. This allows enterprises to build custom routers that intelligently direct requests to the most suitable model. The bottom line is that these new releases aim to make AI more accessible and customizable for specialized tasks.