Top AI Product

NVIDIA debuted the Nemotron 3 family of open models — Nano, Super, and Ultra — positioned as the most efficient open models for building agentic AI applications. The headline: Nemotron 3 Nano delivers 4x higher throughput than Nemotron 2 Nano, and the most tokens per second for multi-agent systems at scale.

## The architecture

Nano’s throughput gain comes from a hybrid mixture-of-experts architecture — activating only a fraction of parameters per token, so you get large-model capability at small-model inference cost. For multi-agent systems where dozens of agents each make many calls, tokens-per-second is the binding constraint, and Nano optimizes exactly that.

## The Omni variant

Nemotron 3 Nano Omni folds vision, audio, and language into a single open model (30B parameters, 3B active) aimed at edge AI agents — unified multimodal reasoning without stitching separate models together. NVIDIA claims up to 9x more efficient agents.

## Why it matters

Open weights plus agentic-throughput optimization is a deliberate combination. NVIDIA wants the open ecosystem building multi-agent systems on models tuned for its hardware — capturing the inference layer regardless of which lab wins the frontier-model race. Nemotron 3 is infrastructure positioning as much as a model release.