Origin
NVIDIA, a US company, was founded in 1993 by Jensen Huang, Chris Malachowsky and Curtis Priem, and is headquartered in Santa Clara, California. It is best known for graphics processing units, a business Wikipedia says has broadened into AI and supercomputing. Its Nemotron research site gives the Nemotron project two goals: to help NVIDIA build systems for building and deploying AI, and to contribute datasets, techniques and models openly to the community.
How it got here
In October 2024 NVIDIA introduced NVLM 1.0, a family of multimodal language models with a 72B flagship, according to Wikipedia. In April 2025 it released the reasoning model Llama-3.1-Nemotron-Ultra-253B-v1 under the NVIDIA Open Model License; the Llama-Nemotron family came in Nano, Super and Ultra sizes. The Parakeet TDT 0.6B v3 speech recognition model, covering 25 European languages, appeared in August 2025.
On 15 December 2025 NVIDIA announced Nemotron 3, a hybrid mixture-of-experts family with Nano, Super and Ultra sizes of roughly 30 billion, 100 billion and 500 billion parameters. Nano went out through Hugging Face that day, according to its model card. Nemotron 3 Super (120B total, 12B active) shipped on 11 March 2026, and Nemotron 3 Ultra (550B total, 55B active) on 4 June 2026. A Nemotron 3.5 Lightning model with 30B total parameters appeared in August 2026.
The Nemotron 3 model cards list NVIDIA’s Nemotron pre-training datasets, which are published on Hugging Face. Some training data ships with the weights.
Licensing pattern
NVIDIA has used several licences, and its newest releases changed again.
Parakeet TDT 0.6B v3 uses CC BY 4.0, according to its card. Llama-3.1-Nemotron-Ultra-253B-v1 used the NVIDIA Open Model License, and its model card adds the Llama 3.1 Community License Agreement as additional information.
Nemotron 3 Nano and Super use the NVIDIA Nemotron Open Model License. The licence page, last modified on 15 December 2025, says the works are commercially usable, that you are free to create and distribute derivative works, and that NVIDIA does not claim ownership of outputs.
Nemotron 3 Ultra and Nemotron 3.5 Lightning moved to OpenMDW 1.1, according to their model cards, which link to the OpenMDW text rather than NVIDIA’s own licence. Adjacent releases can carry different terms, so the licence line on each card matters more than the family name.