Nemotron 3.5 Super

A 120B-parameter multimodal large model from NVIDIA with 12B activated parameters, using a MoE and hybrid Mamba-Transformer architecture, targeting reasoning, long-context understanding, and agentic workloads.