NVIDIA Nemotron 3 Super

Nemotron 3 Super is NVIDIA's open-weight reasoning LLM for agentic workflows, featuring a hybrid Mamba-2 + MoE + attention architecture with approximately 120B total and 12B active parameters per token. It is designed for efficient long-context inference, tool use, and multi-step reasoning in coding, RAG, and IT automation tasks. Source: supabase.