Nemotron-Cascade 2: Open 30B MoE with 3B Active Parameters
NVIDIA's open-weight MoE model hits gold-medal math/coding performance with only 3B active params — practical for agentic subagent routing at low cost.
NVIDIA's open-weight MoE model hits gold-medal math/coding performance with only 3B active params — practical for agentic subagent routing at low cost.