NVIDIA Nemotron 3.5 Lightning: A fast 30B MoE for agent execution
NVIDIA just launched Nemotron 3.5 Lightning, an open 30B MoE model with only 3B active parameters tailored for high-volume agent tasks like tool calling and result validation. It completes 10,000 tasks 30% faster than Qwen3.6 35B on PinchBench. Frontier models handle planning, while Lightning handles the grunt work.