Nvidia

Nemotron 3 Has 120B Parameters and Uses 12B

NVIDIA's new Mixture-of-Experts model activates just 10% of its parameters per query. An order of magnitude cheaper inference changes the ROI calculation for every AI project.