NVIDIA Nemotron 3.5 Lightning and NeMo Switchyard Deliver Faster, Smarter, and More Efficient Agentic AI

NVIDIA has released Nemotron 3.5 Lightning, targeting long-running agentic AI tasks, and the open-source routing library NeMo Switchyard. Both technologies focus on enhancing control over AI deployment and execution across diverse environments such as PCs, workstations, data centers, and the cloud.

Nemotron 3.5 Lightning is an open model based on Mixture-of-Experts (MoE) using 30 billion parameters. It is designed to handle high-frequency specialized tasks in multi-agent systems where multiple models share roles, such as code reviews, tool usage, security alert monitoring, and billing inquiries.

Key features include:

• Up to 4x faster output speed compared to models in the same class
• 30% reduction in agentic task completion time
• Support for post-training based on proprietary data, tools, and workflows using NVIDIA NeMo
• Customization tailored to organizational privacy requirements and deployment locations

NeMo Switchyard is an open-source library that automatically assigns requests to the appropriate model within popular agentic tools. Enterprises can build routers tailored to their own requirements to operate open models, proprietary models, and NVIDIA models together, selecting the most suitable model for each request without rewriting existing applications.

CrowdStrike, Harvey, CodeRabbit, Lila Sciences, and Fastino Labs are applying or customizing Nemotron 3.5 Lightning for specialized agentic tasks in cybersecurity, legal, code review, science, finance, and healthcare.