Nvidia Nemotron 3.5 Lightning

NVIDIA released Nemotron-3.5-Lightning, a 30-billion parameter open-source language model with only 3 billion active parameters, designed for efficient deployment on consumer and enterprise hardware while supporting autonomous agents and local inference. The model uses a hybrid Mixture-of-Experts architecture combining Mamba-2 and attention layers, supports up to 1 million token context length, and is available under the OpenMDW license for commercial use. It achieves strong performance on reasoning benchmarks like GPQA Diamond (75.57%) and MMLU Pro (81.62) while maintaining efficiency through speculative decoding optimizations.