NVIDIA develops the Nemotron family of foundation models.

Nemotron is a line of open LLMs, including Nemotron-4 340B and the Nemotron 3 Nano family, for chat, reward modeling, and synthetic data generation.

Weights are hosted on Hugging Face under the nvidia organization, with training recipes at github.com/NVIDIA-NeMo/Nemotron.

Nemotron-4 340B released in June 2024, with the Nemotron 3 family following in 2026.

Built to give enterprises customizable, commercially permissive open models and synthetic-data tools for building custom LLMs.

Deployable via NVIDIA NeMo, TensorRT-LLM, or Hugging Face Transformers.

Released under the NVIDIA Open Model License, a permissive commercial-use license; weights open, no credentials required to download, self-hosted or via NVIDIA NIM; maintained by NVIDIA.