NVIDIA Open Models: Model List, Licenses and Articles

September 26, 2026September 27, 2026

About NVIDIA

NVIDIA is known as a GPU company, but with 943 models on its nvidia organization on Hugging Face it is also a major publisher of open models. The organization card calls it “NVIDIA’s home for open model weights, datasets, and interactive demos".

NVIDIA’s releases come in two different kinds.

1. Its own models (Nemotron and others). A family of foundation models for language, reasoning, vision, speech and safety. According to the organization card, it ships open weights together with open datasets and reproducible training recipes.
2. Converted builds of other companies’ models. NVIDIA publishes many models from DeepSeek, Qwen, GLM and others quantized to NVFP4, a 4-bit format for its newest GPUs (Blackwell). In the table below, these are the rows with -NVFP4 after another company’s model name.

What It Releases

• The Nemotron 3 language models are MoE models that combine Mamba-2 and Transformer layers. The organization card lists the small Nemotron 3.5 Lightning (30B total, 3B active), Nemotron 3 Super (120B total, 12B active) and the largest, Nemotron 3 Ultra (550B total, 55B active). The Nemotron 3.5 Lightning model card says it supports up to a one-million-token context and a reasoning mode that can be switched on and off.
• It also releases speech recognition and translation, vision-language and safety models (Nemotron Speech, Nemotron 3.5 Content Safety and others).

Licenses

• Many of its own recent models use the OpenMDW License 1.1. The main conditions are keeping the license text and copyright notices when distributing, and losing the rights if you sue claiming the model infringes a patent or copyright; it places no restrictions on the use of outputs.
• The NVIDIA Open Model License is also used. NVIDIA states that models under it are commercially usable, that you are free to create and distribute derivative models, and that it does not claim ownership of outputs.
• NVFP4 builds of other companies’ models either keep the original model’s license or use the NVIDIA Open Model License. Check the license column in the table. Some models use an evaluation-only license (nvidia-software-and-model-evaluation-license).
• This summary is not legal advice; read each model’s license text before use.

Running Them Locally

• NVFP4 builds run fast on Blackwell-generation GPUs. NVFP4 is a format for the 4-bit arithmetic of Blackwell (GB200, B200, the RTX 50 series and others).
• The recommended setups are data-center class. The single-GPU setup assumed by the Nemotron 3.5 Lightning (BF16) model card is one 80GB H100 or A100.
• There are almost no official GGUF builds. The only GGUF we find on the nvidia organization is NVIDIA-Nemotron-3-Nano-4B-GGUF (as of 2026-09-26).

Sources: the nvidia organization card; the model card of nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16; the OpenMDW License 1.1; the NVIDIA Open Model License; the model list of the nvidia organization on Hugging Face (all as of 2026-09-26).

Our Coverage and Data

Local Model Watch has published 1 article(s) on NVIDIA’s own models, 0 on third-party fine-tunes and quantized builds based on them, and 8 from its official blog. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.

At a Glance

Recent Models on Hugging Face

The publisher lists 943 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.

From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.

Licenses of Its Own Models (as recorded in our articles)

The license can differ from model to model even within one publisher. This is not legal advice; check each model’s license text before use.

By Model Family

Articles on Its Own Models

From Its Official Blog

• 2026-09-25 NVIDIA Optimizes MoE Training for Biological Foundation Models
• 2026-09-24 NVIDIA Releases SWE-Serve Benchmark for AI Coding Agents
• 2026-09-23 NVIDIA Topograph: Open Source Cluster Topology Toolkit
• 2026-09-23 AI Inference Performance with NVIDIA Blackwell Confidential Computing
• 2026-09-22 NVIDIA Dynamo-Triton 26.07 Adds Multi-Device Inference
• 2026-09-19 NVIDIA AIPerf: Benchmarking LLM Inference at Scale
• 2026-09-15 NVIDIA Accelerates Dropless MoE Training in JAX

Last updated 2026-09-27 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.