Skip to content

Nemotron models

NVIDIA's open models with open weights, data and recipes: hybrid Mamba-Transformer MoE models with up to a 1M-token context.

  • Open weights
  • Models 7
  • Context up to 1M
  • Updated Nov 2026
Hugging Face Compare
On this page

At a glance

Type
Open-weight models
Developer
NVIDIA (United States)
First release
Jun 2024
Models tracked
7 (5 current)
Longest context
1M tokens
Abilities
Tool callingReasoning modeImage inputAudio input
Licence
Nemotron 3.5 and Nemotron 3 Ultra use OpenMDW 1.1, a permissive open model licence. Older Nemotron 3 models use the NVIDIA Nemotron Open Model License, which also allows commercial use.
Smallest local option
Nemotron 3 Nano 4B · ~4 GB
Website
nvidia.com
Hugging Face
huggingface.co/nvidia

Newest model

Nemotron 3.5 Super

Released Oct 2026 · 1M context

Smallest option

Nemotron 3 Nano 4B

~4 GB VRAM · 2.8 GB download

Nemotron models and versions

ModelReleasedSizesContextAbilitiesLicenceStatus
Nemotron 3.5 Super Vision-language model for coding, tools and documentsOct 2026 120B 120B total, 12B active (MoE)1MToolsReasoningVisionOpenMDW 1.1 Current
Nemotron 3.5 Bolt Compact local model built from Gemma 4 E4BOct 2026 E4B E4B, 74B (9B active)128KToolsReasoningVisionAudioOpenMDW 1.1 Current
Nemotron 3.5 Lightning Fast always-on agent model with 1M contextAug 2026 30B 30B total, 3B active (MoE)1MToolsReasoningOpenMDW 1.1 Current
Nemotron 3 Ultra Largest Nemotron, for long agent workflowsJun 2026 550B 550B total, 55B active (MoE) Too large for most servers · Ollama cloud1MToolsReasoningOpenMDW 1.1 Current
Nemotron Cascade 2 Reasoning and agentic MoEMar 2026 30B 30B total, 3B active (MoE)256KToolsReasoningNVIDIA Open Model License Current
Nemotron 3 Super Text-only predecessor of 3.5 SuperMar 2026 120B 120B total, 12B active (MoE)256KToolsReasoningNVIDIA Nemotron Open Model License Previous generation
Nemotron 3 Nano Small reasoning modelsDec 2025 4B · 30B 4B, 30B (3B active)1MToolsReasoningNVIDIA Nemotron Open Model License Previous generation

Sizes, context windows and abilities come from the Ollama library and Hugging Face model cards, checked 14 Nov 2026. Release months are the first public upload. Spotted a mistake? Tell us.

How much VRAM does Nemotron need?

Download sizes per Ollama build and the GPU memory each needs with an 8K context. On a CPU-only server the same number is the RAM you need. Longer contexts need more memory; work it out with the LLM VRAM calculator.

Model4-bit (default)8-bit16-bitVRAM (4-bit)Fits onCommand
Nemotron 3.5 Super 120B90 GB133 GB249 GB~100 GB128 GB GPU
Nemotron 3.5 Bolt E4B7.9 GB9.8 GB17 GB~9.5 GB12 GB GPU
Nemotron 3.5 Lightning 30B25 GB35 GB66 GB~28.5 GB32 GB GPU
Nemotron Cascade 2 30B24 GB——~27 GB32 GB GPU
Nemotron 3 Super 120B87 GB132 GB247 GB~96.5 GB128 GB GPU
Nemotron 3 Nano 4B2.8 GB4.2 GB8 GB~4 GB8 GB GPU
Nemotron 3 Nano 30B24 GB34 GB63 GB~27 GB32 GB GPU

VRAM = 4-bit download + about 10% runtime overhead + 0.5 GB for an 8K context, rounded up. Mixture-of-experts (MoE) models still need all their weights in memory, but they run much faster than a dense model of the same size, so they also work well split between GPU and system RAM.

Run Nemotron 3.5 Super locally

# Install Ollama (Linux), then pull and chat
curl -fsSL https://ollama.com/install.sh | sh
ollama run nemotron-3.5-super:120b

New to Ollama? Follow Install Ollama on Ubuntu 26.04 (or 24.04) for GPU drivers, the systemd service and remote access.

What Nemotron is good for

Agents with long memory

Lightning and Super handle up to 1M tokens of context for long tool histories.

Fast inference on NVIDIA GPUs

The models are tuned for NVIDIA hardware and NVFP4 4-bit precision.

Training your own model

NVIDIA publishes the training data and recipes, not just the weights.

Strengths and limits

Strengths

  • Open weights, data and recipes.
  • Hybrid Mamba-Transformer design keeps long contexts cheap.
  • Permissive OpenMDW licence on the newest models.

Watch out for

  • Best performance assumes recent NVIDIA GPUs.
  • Smaller community than Qwen or Llama.

About Nemotron

Nemotron is NVIDIA’s family of open models. Early versions were fine-tunes of Llama; since Nemotron 3 (December 2025) NVIDIA trains its own hybrid Mamba-2/Transformer mixture-of-experts models from scratch and publishes the weights, training data and recipes.

The 2026 line-up spans Nemotron 3.5 Bolt (a compact vision model post-trained from Gemma 4 E4B), Nemotron 3.5 Lightning (30B total, 3B active, 1M context) for always-on agents, Nemotron 3.5 Super (120B, 12B active, with vision) and Nemotron 3 Ultra (550B, 55B active). The 3.5 generation moved to OpenMDW 1.1, a permissive licence for open models.

Guides for running Nemotron

Tool LLM VRAM Calculator: How Much GPU Memory Do You Need? Updated Nov 2026 AI & LLMs How to Install Ollama on Ubuntu 26.04 and Run LLMs Locally Updated Nov 2026 AI & LLMs How to Install llama.cpp on Ubuntu 26.04 (CPU and NVIDIA GPU) Updated Nov 2026 AI & LLMs How to Install vLLM on Ubuntu 26.04 (OpenAI-Compatible LLM Server) Updated Nov 2026 AI & LLMs How to Install Open WebUI on Ubuntu 26.04 (Self-Hosted ChatGPT Alternative) Updated Nov 2026

Frequently asked questions

What is the latest Nemotron model?

The newest Nemotron model is Nemotron 3.5 Super, released in Oct 2026. It has a 1M-token context window and supports tool calling, reasoning mode, image input.

How much VRAM do I need to run Nemotron locally?

The smallest option, Nemotron 3 Nano 4B, is a 2.8 GB download at 4-bit and needs about 4 GB of GPU memory (or RAM on a CPU-only machine) with an 8K context. Larger sizes need more; see the hardware table on this page, and use the VRAM calculator for longer contexts.

How do I run Nemotron with Ollama?

Install Ollama, then run "ollama run nemotron-3-nano:4b". Ollama downloads the 4-bit build and starts a chat; the same model is also available on the local API at http://localhost:11434.

Can I use Nemotron commercially?

Nemotron 3.5 and Nemotron 3 Ultra use OpenMDW 1.1, a permissive open model licence. Older Nemotron 3 models use the NVIDIA Nemotron Open Model License, which also allows commercial use. Always read the licence of the exact model you deploy; it is linked in the models table.

Compare Nemotron

← All AI models

Back to top