Nvidia Launches Its Open Model Family Nemotron 3 for Artificial Intelligence
American chip manufacturer Nvidia announced on December 15 the launch of Nemotron 3, a family of open-access artificial intelligence models, accompanied by training data and software libraries.
Diverse Model Sizes within Nemotron 3
According to the company, the Nemotron 3 family consists of three models of distinct sizes: Nano, with 30 billion parameters; Super, approximately 100 billion parameters; and Ultra, reaching nearly 500 billion parameters. The Nemotron 3 Nano model is immediately available as an NVIDIA NIM microservice, while the Super and Ultra versions are expected in the first half of 2026. The Nano model activates 3.2 billion parameters out of a total of 31.6 billion. The company states that Nemotron 3 Nano offers a throughput that is four times higher than its predecessor, Nemotron 2, and reduces reasoning token generation by up to 60%. The models support a context length of up to 1 million tokens.
Open Models Designed for Sector-Specific AI Development
Nvidia states that these open models, data, and libraries are designed to fuel the development of transparent, efficient, and specialized agent-based artificial intelligence across various sectors. The Nemotron 3 models introduce a hybrid architecture of latent expert mixing that assists developers in building and deploying reliable large-scale multi-agent systems. All model weights, training corpora, and detailed recipes are available on GitHub and Hugging Face under the NVIDIA Open Model License, and developers also have access to the open-source libraries NeMo Gym, NeMo RL, and NeMo Evaluator. The group has also released a collection of training datasets and reinforcement learning libraries.
Widespread Adoption of Nemotron Models
According to the press release, several companies are already adopting the models from the Nemotron family, including Accenture, Cadence, CrowdStrike, Cursor, Deloitte, EY, Oracle Cloud Infrastructure, Palantir, Perplexity, ServiceNow, Siemens, Synopsys, and Zoom, to power artificial intelligence workflows in sectors such as manufacturing, cybersecurity, software development, media, communications, and other industries. Portfolio companies of General Catalyst and Mayfield are also exploring Nemotron 3 for building AI teams that support human-machine collaboration. Nvidia emphasizes that Nemotron is part of its broader efforts in sovereign artificial intelligence, enabling organizations from different countries to build systems aligned with their own data, regulations, and values.
