FrontierAugust 12, 2026via MarkTechPost
NVIDIA AI Releases Nemotron 3.5 Lightning: A 30B Open MoE with 3B Active Parameters, and NeMo Switchyard Model Router
Why it matters
A major open-weight release targeting agent execution efficiency. Nemotron 3.5 Lightning and Switchyard together signal NVIDIA's play to own the inference layer for agentic workloads—competing directly with smaller, cost-optimized models from rivals.
Key signals
- Nemotron 3.5 Lightning: 30B parameters, 3B active (MoE architecture)
- Open-weight release (no fence)
- NeMo Switchyard: model router for agent execution layer
- Routing strategy: selects cheapest capable model per step
- Agent execution framed as primary use case
- Published August 11, 2026
The hook
NVIDIA open-sources a 30B MoE with just 3B active parameters—and ships a router that picks the cheapest capable model for each step.
NVIDIA's open 30B MoE targets the agent execution layer, with Switchyard routing each step to the cheapest capable model.