Details
- NVIDIA introduced Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts open model aimed at high-volume agentic workloads and specialized tasks inside multi-agent systems.
- The company also released NeMo Switchyard, an open-source Rust routing library that sits between agent tools and model backends to route requests without changing application code.
- Nemotron 3.5 Lightning is designed for customizable deployment across local systems, PCs, workstations, data centers and cloud environments, including RTX PCs, DGX systems and Jetson devices.
- Switchyard supports protocol translation and multi-backend routing across OpenAI- and Anthropic-compatible APIs, with configurable algorithms for quality, latency and cost tradeoffs.
- NVIDIA says internal benchmarks and partner tests show lower task costs and faster completion, while the Switchyard repository currently labels the software experimental and pre-alpha.
Impact
The release reinforces a broader shift toward agent systems built from ensembles of specialized models rather than a single default model. If the benchmarks hold in real deployments, it could accelerate enterprise adoption of open, locally deployable models and increase demand for routing layers that optimize cost and latency across heterogeneous model stacks.
