Teams building multi-agent systems get a free open family with training data included; hosted pricing depends on the provider.
Kenning's take
What it does
Nemotron 3 is a family of open models from NVIDIA aimed at multi-agent AI systems, released with open training data and reinforcement learning tools. NVIDIA announced it in December 2025 with Nemotron 3 Nano (30B parameters, about 3B active) available immediately, followed by the larger Super and Ultra models in 2026. The weights are on Hugging Face and the models run through providers such as OpenRouter, Together AI and Amazon Bedrock.
- Nano, Super and Ultra sizes using mixture-of-experts
- Open weights, training datasets and RL environments
- Up to 1 million token context window
- Supported by vLLM, SGLang, llama.cpp and LM Studio
- Packaged as NVIDIA NIM microservices
Who it suits. Developers and enterprises building agents who want open, efficient models they can customise.
Key facts
- Category
- AI models
- Made by
- NVIDIA
- Based in
- United States
- Released
- 15 December 2025
- How to get it
- open weights on Hugging Face; hosted by providers including OpenRouter, Together AI, Fireworks, DeepInfra and Amazon Bedrock, and on build.nvidia.com
- Works on
- API, self-hosted
- In the UK
- Yes
- Free plan
- free open weights
- API price
- set by third-party providers; NVIDIA does not publish one price
Models and features
- Nemotron 3 Nano Omni 2026-04-28
- Nemotron 3 Super 2026-03-11
- Nemotron 3.5 Lightning 2026-08-11
- NVIDIA Nemotron 3 Ultra 2026-06-04
FAQ
When was Nemotron 3 released?
NVIDIA announced the Nemotron 3 family on 15 December 2025, with Nano available that day.
Is Nemotron 3 open source?
NVIDIA released the weights, training data and reinforcement learning libraries openly, on Hugging Face and GitHub.
What sizes does Nemotron 3 come in?
Nano (30B parameters, about 3B active), Super and Ultra.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.