NVIDIA ships Nemotron 3.5 Lightning and NeMo Switchyard router
NVIDIA released NeMo Switchyard, an open-source routing library on GitHub that sends each step of an agent workflow to the best-fit model across open, proprietary and NVIDIA models without any application rewrite, alongside Nemotron 3.5 Lightning, a 30B mixture-of-experts model built for high-volume agent tasks. Internal benchmarks held frontier-level accuracy at nearly one-third the task-completion cost of Opus 4.8 alone, and LangChain measured 74% lower cost across 145 multi-turn Deep Agents tasks by routing 7% of calls to a frontier model at a 6% accuracy tradeoff. A developer running LangChain, LiteLLM or Kong can add per-step routing today; the tradeoff is that every published cost number is vendor-run, with adopters including Cognition, Ramp and Siemens.
Source: blogs.nvidia.com ↗
NeMo Switchyard maintains frontier-level accuracy while reducing task completion cost to nearly one-third of Opus 4.8 alone.
Why this matters
- → Agents can now route tasks to cheaper models without rewriting code, cutting costs 58-74% vs frontier-only.
- → Open 30B model runs locally for privacy-critical specialized tasks like code review and security monitoring.
- → System-of-models architecture shifts control from vendor lock-in to developer choice across open and proprieta