Nvidia releases first open-source AI model Nemotron 3.5 Lightning

Nvidia launched Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters, 1M context, and strong agentic benchmarks (SWE-bench 52.8%). It is free to modify and runs on a single GPU, signaling Nvidia's strategic pivot to open-weight models. Additionally, Nvidia introduced NeMo Switchyard routing library to optimize token costs by routing to cheaper models, with enterprise partners Ramp, LangChain, Kong.