TellWorldAI World Archive
Ask中文
News
Models
Organisations
People
Papers & concepts
Products & agents
Chips & robots
History
Topics
Ask中文

Home›Companies

OrganizationResearch lab

NVIDIA Research

NVIDIA Research is the research arm of the chipmaker NVIDIA, founded in 2006 with more than 200 scientists. It leads work on AI, self-driving cars, high-performance computing and graphics, and it builds the company's open Nemotron language models.

Founded 2006Website research.nvidia.com Era Statistical learning

2025 · 5 20212026
Events per year
12Models & products10Events1Investors & partners

What it is

NVIDIA Research is the research organisation of NVIDIA, the company best known for the graphics processing units (GPUs) that now run most AI systems. A GPU is a chip built to carry out many calculations at the same time, which suits the arithmetic of neural networks. The research group works on AI, self-driving cars, high-performance computing, graphics, and virtual and augmented reality. It has more than 200 scientists [1].

The parent company, NVIDIA, was founded on 5 April 1993 by Jensen Huang, Chris Malachowsky and Curtis Priem, and is headquartered in Santa Clara, California [2].

Origins

NVIDIA Research was founded in 2006 by David Kirk, who was then NVIDIA's chief scientist [1]. A chief scientist is the senior technical leader of a company, and in this case the role linked the research group directly to the company's technical direction. Bill Dally joined from Stanford in 2009 [1]. Applied deep-learning research is led by Bryan Catanzaro [1].

Key features

The group's work covers several fields at once, which is part of its character. The AI work includes language and vision models, robotics and self-driving research. The high-performance computing and graphics work is part of the same portfolio [1].

Its open model lines are among the most visible results. The Nemotron language models include Nemotron-3 8B in November 2023, Nemotron-4 340B in June 2024 and the Llama Nemotron reasoning models in 2025 [5][2]. The Nemotron 3 family, which ran from December 2025 to June 2026, uses hybrid Mamba-Transformer mixture-of-experts designs. NVIDIA released the weights, training data and recipes for these models [4][5]. A mixture-of-experts design sends each input to a few specialist parts of the network, and Mamba is a different sequence architecture that is combined here with the transformer.

Other open lines include NVLM multimodal models, released in 2024, which handle both images and text [2]. Cosmos world models, announced in January 2025, are a platform for physical AI, meaning AI that works with the physical world [6][2]. Isaac GR00T robot models are part of the group's robotics work [2]. Alpamayo-R1, released on 1 December 2025, is a vision-language-action model for self-driving [2].

Research at the group also contributes to large training projects. With Microsoft it trained Megatron-Turing NLG 530B, announced on 11 October 2021, on NVIDIA's Selene supercomputer using the Megatron-LM and DeepSpeed libraries [3].

Where it sits

In March 2026 NVIDIA launched the Nemotron Coalition. Its first project is a base model co-developed with Mistral AI for a future Nemotron 4 family [5]. A base model is a general model that other work can build on. Nemotron 3.5 models followed from August 2026, including Super and Lightning (14 October 2026) [4][5].

For readers, NVIDIA Research is a useful way to see how a hardware company also builds models. Its Nemotron 3 releases include the weights, training data and recipes, which makes the work easier to check and reuse.

TechniquesFoundation modelsSimulation

ApplicationsRobotics

Ecosystem & regionOpen sourceHardware & compute

20261

  1. ReleaseNVIDIA releases Nemotron 3 Ultra, a 550B open model

    On 4 June 2026 NVIDIA released Nemotron 3 Ultra, the largest model in its Nemotron 3 family, with 550 billion total parameters and 55 billion active per token.

20255

  1. ReleaseNVIDIA debuts the Nemotron 3 family of open models

    On 15 December 2025 NVIDIA announced the Nemotron 3 family of open models, releasing Nano at once and promising Super and Ultra in 2026 along with three trillion tokens of data.

  2. ReleaseMistral AI releases the Mistral 3 family, including Mistral Large 3

    On 2 December 2025 Mistral AI released the Mistral 3 family: Mistral Large 3, a 675B mixture-of-experts model, and three Ministral 3 models, all under Apache 2.0.

  3. MilestoneMistral AI announces Mistral Compute with NVIDIA

    On 11 June 2025 Mistral AI announced Mistral Compute, an AI infrastructure offering that sells GPUs, orchestration, APIs and services as bare metal or managed PaaS, built with NVIDIA hardware and reference architectures.

  4. ReleaseNVIDIA launches the Llama Nemotron open reasoning models

    On 18 March 2025, NVIDIA launched the Llama Nemotron family of open reasoning models at its GTC conference, with partners including Microsoft and SAP.

  5. ReleaseNVIDIA launches the Cosmos world foundation model platform at CES

    On 6 January 2025, at CES, NVIDIA launched Cosmos, a platform of open world foundation models for physical AI used to train robots and autonomous vehicles.

20243

  1. ReleaseMistral AI and NVIDIA release Mistral NeMo 12B

    On 18 July 2024 Mistral AI released Mistral NeMo, a 12B language model built with NVIDIA, with a 128K-token context window, a new multilingual tokenizer called Tekken, FP8-aware training and Apache 2.0 weights.

  2. ReleaseNVIDIA releases Nemotron-4 340B for synthetic data generation

    On 14 June 2024 NVIDIA released Nemotron-4 340B, a family of base, instruct and reward models built for generating synthetic training data. It published the models under a permissive license, and the reward model topped RewardBench.

  3. ReleaseNVIDIA announces Project GR00T

    In March 2024 NVIDIA announced Project GR00T, a general-purpose foundation model for humanoid robots. A year later it released Isaac GR00T N1, which it presented as a fully customizable humanoid model.

20211

  1. ReleaseNVIDIA and Microsoft unveil the 530-billion-parameter Megatron-Turing NLG

    On 11 October 2021 NVIDIA and Microsoft announced Megatron-Turing NLG 530B, a language model with 530 billion parameters that they called the largest and most powerful of its kind. The full technical report followed in January 2022.

Models and products over time

11
  1. 2021
    • Oct 11, 2021 Megatron-Turing NLG 530B Model

      Megatron-Turing NLG 530B is a 530-billion-parameter language model announced by NVIDIA and Microsoft on 11 October 2021, then the largest monolithic transformer language model, trained on NVIDIA's Selene supercomputer with Megatron-LM and DeepSpeed combined.

  2. 2024
    • Mar 18, 2024 Project GR00T Model

      NVIDIA's project for a general-purpose foundation model for humanoid robots, announced in March 2024 and followed in March 2025 by Isaac GR00T N1, an open, customizable model released with training data and evaluation scenarios.

    • Jun 14, 2024 Nemotron-4 340B Model

      Nemotron-4 340B is a family of open models from NVIDIA, released on 14 June 2024 as base, instruct and reward models. It was built to generate synthetic data for training other language models, and it fits on a single eight-GPU DGX H100.

    • Jul 18, 2024 Mistral NeMo Model

      Mistral NeMo is a 12-billion-parameter language model released by Mistral AI with NVIDIA in July 2024, notable for a 128K-token context window, a new multilingual tokenizer called Tekken, and an Apache 2.0 release.

    • Sep 17, 2024 NVLM 1.0 Model

      NVLM is a family of multimodal large language models from NVIDIA, described in a paper posted to arXiv in September 2024; its flagship NVLM-D-72B was later released as open weights for non-commercial use.

    • Oct 2024 Llama-3.1-Nemotron-70B-Instruct Model

      NVIDIA's Llama-3.1-Nemotron-70B-Instruct is a version of Meta's Llama-3.1-70B-Instruct tuned with reinforcement learning from human feedback to give more helpful answers; on release it topped three automatic alignment benchmarks.

  3. 2025
    • Jan 6, 2025 NVIDIA Cosmos world foundation models Model

      NVIDIA Cosmos is a platform of world foundation models that generate physics-based video to train robots and autonomous vehicles, released as open-weight models in January 2025.

    • Mar 18, 2025 Llama Nemotron (Nano, Super, Ultra) Model

      Llama Nemotron is NVIDIA's family of open reasoning models in Nano, Super and Ultra sizes, launched in March 2025 and built on Meta's Llama 3 models.

    • Aug 20, 2025 NVIDIA Nemotron Nano 2 Model

      Nemotron Nano 2 is NVIDIA's 9B reasoning model, built on a hybrid Mamba-Transformer design and released with open weights and most of its training data; NVIDIA reports much higher throughput than similar-sized models.

    • Dec 2, 2025 Mistral Large 3 Model

      Mistral AI's flagship model of the Mistral 3 family, released in December 2025: a sparse mixture-of-experts model with 675B total parameters, open weights under Apache 2.0, and image understanding.

    • Dec 15, 2025 NVIDIA Nemotron 3 (Nano, Super, Ultra) Model

      NVIDIA Nemotron 3 is an open family of hybrid mixture-of-experts language models, announced in December 2025 in Nano, Super and Ultra sizes, with up to one million tokens of context.

Part of1

NVIDIA

Notes and sources

  1. About NVIDIA Researchhttps://www.nvidia.com/en-us/research/about
  2. Nvidiahttps://en.wikipedia.org/wiki/Nvidia
  3. Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530Bhttps://developer.nvidia.com/blog/using-deepspeed-and-megatron-to-train-megatron-turing-nlg-530b-the-worlds-largest-and-most-powerful-generative-language-model/
  4. Nemotron projectshttps://research.nvidia.com/labs/nemotron/projects/
  5. Nemotronhttps://en.wikipedia.org/wiki/Nemotron
  6. NVIDIA Launches Cosmos World Foundation Model Platform to Accelerate Physical AI Developmenthttps://nvidianews.nvidia.com/news/nvidia-launches-cosmos-world-foundation-model-platform-to-accelerate-physical-ai-development

Cite this page: NVIDIA Research. TellWorld, https://www.tellworld.net/companies/nvidia-research. Updated 2026-11-08