Skip to content

ModelsNVIDIAUpdated

NVIDIA debuts the Nemotron 3 family of open models for agentic AI

Nemotron 3 Nano at a glance

NVIDIA announced Nemotron 3 in Nano, Super, and Ultra sizes, shipping the 30B Nano with a 1M-token context and open training data on day one.

What changed with Nemotron 3 Nano

On December 15, 2025, NVIDIA announced the Nemotron 3 family: Nano (30B total, up to 3B active), Super (about 100B, up to 10B active), and Ultra (about 500B, up to 50B active). Nano was available that day on Hugging Face and several inference providers; Super and Ultra were slated for the first half of 2026.

NVIDIA says Nano has a 1M-token context, up to 4x the token throughput of Nemotron 2 Nano, and up to 60% fewer reasoning tokens. The release also included open pretraining and post-training datasets, NeMo Gym and NeMo RL libraries, and an agent safety dataset.

Nemotron 3 Nano in numbers

Values are copied from the linked source on the date shown. They are not IA Compa measurements.

Context windowas of · Source for Context window: nvidianews.nvidia.com (opens in a new tab)
1M tokens (Nano)

How it compares

Other models in the same category that we have covered, using each model's latest recorded value. (Models)

Context window (tokens) · Nemotron 3 Nanohigher is better

Where to compare Nemotron 3 Nano

Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.

Why Nemotron 3 Nano matters for businesses

Small, efficient open models like Nano are a good fit for multi-step agents where cost per call matters, and open datasets make it easier for teams to audit or fine-tune for a niche domain.

Nemotron 3 Nano updates since publication

Dated changes since publication. The original article text above is kept as written on its date.

  1. Correction (Updated)

    Nemotron 3 Ultra shipped on June 4, 2026 at 550B total parameters with about 55B active (NVIDIA-Nemotron-3-Ultra-550B-A55B), larger than the roughly 500B announced in December.

    huggingface.co (opens in a new tab)

Sources for this story

  1. Official

Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.

  • #llm
  • #nvidia
  • #nemotron
  • #open-weights
  • #agents

Want this working in your business?

We turn new AI capabilities into agents, voice receptionists, and automations you can actually use.