Nvidia ships Nemotron 3.5 Lightning

Nvidia released Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts model built for specialized, high-volume tasks inside multi-agent systems, such as code review, security monitoring, and billing inquiries. Nvidia claims up to 4x faster output and 30% faster agentic task completion than comparable models, while keeping frontier-level accuracy at what it says is nearly a third of the cost of running Opus 4.8 alone, when paired with its new NeMo Switchyard routing layer.

Switchyard is an open-source library for routing requests across mixed model ecosystems without rewriting the calling application. The model is available on Hugging Face, ModelScope, and OpenRouter, as a NIM microservice on build.nvidia.com, and through Nvidia Cloud Partners; it can also be post-trained with NeMo on an organization’s own data.

What it means for you

The pitch here isn’t “replace your frontier model,” it’s “stop paying frontier prices for the 80% of agent calls that don’t need frontier reasoning.” If your agent stack routes every subtask through the same large model regardless of difficulty, Nemotron 3.5 Lightning plus Switchyard is a concrete test case for the kind of tiered routing covered in choosing how hard your AI thinks and model price war routing guide. Treat Nvidia’s own cost and speed numbers as vendor claims to verify against your own workload before committing, not as settled fact.