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

NVIDIA launched Nemotron 3 Super — a 120B total / 12B active parameter hybrid Mamba-Transformer MoE, open, designed specifically for multi-agent workloads, and delivering 5x higher throughput than Nemotron 2 at the same active parameter count (NVIDIA Newsroom). It ships with a 1M-token context window targeting complex tasks: software development, cybersecurity triage, multi-agent coordination.

The early adopter list tells the story: Cursor, CrowdStrike, Palantir, Perplexity, and ServiceNow. When an IDE, a cybersecurity firm, a defense contractor, an AI search engine, and an enterprise platform all adopt the same open model backbone simultaneously, that's not a press release — it's an infrastructure decision.

The companion Nemotron 3 Nano (30B total / 3B active) is available now for DGX Spark, H100, and B200, delivering 4x throughput over Nano 2 (NVIDIA Newsroom). Ultra (highest-complexity reasoning) is coming H1 2026. NVIDIA is building the full open model stack for enterprise verticals where proprietary models face regulatory or latency constraints — and the Mamba-Transformer hybrid architecture means this isn't just a bigger transformer, it's a genuinely different inference profile.