NVIDIA Expands Nemotron, Introduces Switchyard Router
NVIDIA rolled out Nemotron 3.5 Lightning as a lightweight, open 30‑billion-parameter mixture-of-experts model, and introduced NeMo Switchyard, an open-source router that directs an agent’s workflow to the most suitable model to cut costs. The router is described as transforming an agent stack into a dispatch system with policies, fallbacks, and logging to explain cost versus performance. Early benchmarks reportedly show cost reductions when routing between Nemotron 3.5 Lightning and larger models like Opus 4.8, though figures are vendor-sourced and should be treated cautiously. Reports indicate NVIDIA is developing Nemotron 4, potentially exceeding 1 trillion parameters, and aims to build an ecosystem of open models optimized for NVIDIA GPUs with a cloud-budget cap of $7 billion through 2028. This broader push aligns with CEO Jensen Huang’s defense of open AI and the belief that free/open models accelerate hardware adoption and innovation. Nemotron 3.5 Lightning remains freely downloadable for code review, billing questions, and security monitoring, with no Nemotron 4 release date announced yet.
Where do you stand?
How it spread
Claim check
What each side asserts, disputes — or leaves out entirely.
Where do you land?
Whose framing of this story rings truest to you?
How each side headlines it
Center· 2 sources
“Nvidia Develops New Open Source AI Model Family, Says Website”Right· 1 source

