Nemotron is a family of artificial intelligence models developed by Nvidia. It includes large language models and multimodal models intended for reasoning, computer programming, information retrieval and agentic AI applications. Nvidia has released open model weights, training data, software and training methods for parts of the family.
History Nvidia introduced the Nemotron-3 8B models in November 2023 for enterprise generative artificial intelligence applications. Nemotron-4 340B followed in June 2024 and included base, instruction-tuned and reward models designed partly for generating synthetic data.
Nvidia later introduced Llama Nemotron, a series of reasoning models derived from Meta's Llama models. In December 2025, Nvidia announced the Nemotron 3 generation, beginning with Nemotron 3 Nano. Nemotron 3 Super and Ultra were released in 2026.
Models + Major model families Family Released Purpose Nemotron-3 2023 Enterprise applications Nemotron-4 2024 Synthetic-data generation Llama Nemotron 2025 Llama-based reasoning Nemotron 3 2025 Agentic AI
The Nemotron 3 models use a hybrid architecture combining Mamba, Transformer, and mixture of experts components.
+ Nemotron 3 models Model Parameters Intended use Nano 30B; 3B active Efficient agents Super 120B; 12B active Agentic reasoning Ultra 550B; 55B active Complex reasoning Nano Omni 30B; 3B active Multimodal agents
Nano Omni supports text, image, video and audio input. Nemotron models can be downloaded for local deployment or accessed through Nvidia's application programming interfaces and NIM inference services.
See also List of AI-assisted software development tools List of large language models Lists of open-source artificial intelligence software Open weights
References
External links
Nvidia models at Hugging Face