NVIDIA Nemotron 3 Super Launches: 120B/12B Active MoE for Agentic AI at 5x Prior Throughput

NVIDIA Newsroom·high signal

NVIDIA launched Nemotron 3 Super today, a 120B total / 12B active parameter hybrid Mamba-Transformer MoE open model designed specifically for multi-agent and agentic workloads, delivering 5x higher throughput than Nemotron 2 at the same active parameter count. The model targets complex tasks like software development and cybersecurity triage with a 1M-token context window. Early adopters include Cursor, CrowdStrike, Palantir, Perplexity, and ServiceNow — a signal that this could become the default open backbone for enterprise agent deployments.