← CookbookView source on GitHub ↗

NVIDIA Nemotron Models

Browse NVIDIA Nemotron models available on Nebius Token Factory.


Available Nemotron Models

Model Provider Parameters Context Highlights
Nemotron-3.5-Lightning

▶ Try it @ TF
NVIDIA 30B total
3B active
1 M Fast, low-cost MoE model built as the execution layer for always-on, long-running agents and high-throughput specialized tasks
Nemotron-3-Super-120B-A12B

▶ Try it @ TF
NVIDIA 120B total
12B active
256 K hybrid MoE model optimized for efficient multi-agent AI and complex reasoning tasks.
Nemotron-3-Ultra-550B-A55B

▶ Try it @ TF
NVIDIA 550B total
55B active
1 M Flagship hybrid MoE model optimized for the most demanding multi-agent AI and complex reasoning tasks

Running the Notebooks

This directory is a uv project. From the repo root:

cd models/nemotron
uv sync

Create a .env file with your Nebius API key (get one at tokenfactory.nebius.com):

cp env.example .env
# then edit .env and paste your key

Start Jupyter and open a notebook:

uv run jupyter lab run_nemotron.ipynb
Notebook Model Runs on
run_nemotron.ipynb nvidia/Nemotron-3_5-Lightning Local (minimal)
run_nemotron_colab.ipynb nvidia/Nemotron-3_5-Lightning Google Colab or local