NVIDIA Open Model License
The NVIDIA Open Model License (in full, the NVIDIA Open Model License Agreement) is a custom license that NVIDIA wrote for the trained weights and related files of the AI models it publishes for download. It allows commercial use, the creation and distribution of derivative models, and use of the outputs, which NVIDIA says it does not own, subject to conditions that standard software licenses do not contain: the license ends if the user circumvents a model's safety guardrails or sues claiming that the model infringes a copyright or patent, and use must be consistent with NVIDIA's Trustworthy AI terms.[1] NVIDIA first published the agreement on June 14, 2024 with the Nemotron-4 340B models[2][3] and revised it five times between January and October 2025; the current version as of November 2026 is dated October 24, 2025.[1][5][6][7][8] It is the license most often named on NVIDIA's open weights releases and covers language, world, robotics, weather and speech models, including Nemotron, Cosmos, Isaac GR00T, Earth-2 and Parakeet checkpoints.[28]
NVIDIA has since issued related licenses. The NVIDIA Nemotron Open Model License (December 15, 2025) and the NVIDIA Open Model Agreement (April 2, 2026) are texts modeled more closely on the Apache License that drop the guardrail and Trustworthy AI conditions,[11][12] while research checkpoints use separate non-commercial licenses.[13][14] In May 2026 NVIDIA said it would move future releases of its Cosmos, Isaac GR00T, Ising and Nemotron families to the Linux Foundation's OpenMDW license,[23] although as of November 2026 new repositories under the NVIDIA Open Model License were still appearing on its Hugging Face account.[28] None of NVIDIA's model licenses is on the Open Source Initiative's list of approved licenses or on the SPDX License List.[50][35]
Background
Before the Open Model License, NVIDIA distributed downloadable language models such as the Nemotron-3 8B family and the Retro 48B and 8B instruct models (November and December 2023) under the NVIDIA AI Foundation Models Community License Agreement of November 10, 2023.[17][28] That agreement allowed installation, modification and distribution only for "non-production uses, including test and evaluation uses"; production use required a paid subscription to the NeMo framework through NVIDIA AI Enterprise. It also barred licensees from using the models, or their outputs, to develop or improve any other large language model.[17]
The Open Model License reversed both restrictions. NVIDIA released Nemotron-4 340B Base, Instruct and Reward on June 14, 2024, and its research page said the models were "open access under the NVIDIA Open Model License Agreement," a permissive license allowing the distribution, modification and use of the models and their outputs.[3] NVIDIA's announcement blog called it "a uniquely permissive open model license" and presented the family as a free way to generate synthetic data for training other models, including commercial ones.[4] The Hugging Face cards for the 340B models still link to the June 2024 text as a PDF.[30]
When NVIDIA launched Cosmos at CES on January 6, 2025, it said developers could "freely use the Cosmos models under NVIDIA's permissive open model license that allows commercial usage," and published a revised agreement the same day that added terms specific to Cosmos.[10][5]
Terms
The summary below follows the version dated October 24, 2025, the text at the agreement's canonical address in November 2026.[1] Section numbers follow NVIDIA's own numbering.
Grant
The agreement opens with three statements that NVIDIA "confirms": models are commercially usable, the licensee is free to create and distribute derivative models, and NVIDIA does not claim ownership of outputs generated with the models or derivative models.[1] The operative grant (section 2.2) is a "perpetual, worldwide, non-exclusive, no-charge, royalty-free, revocable (as stated in Section 2.1) license to publicly perform, publicly display, reproduce, use, create derivative works of, make, have made, sell, offer for sale, distribute (through multiple tiers of distribution) and import the Model."[1] "Model" covers the model itself together with its software, checkpoints, learned weights, algorithms, parameters, configuration files and documentation. A "Derivative Model" is any modification of or work based on the model, and the definition states that "An output is not a Derivative Model."[1]
NVIDIA owns the model and any derivative models it creates; licensees own the derivative models they create, subject to NVIDIA's underlying rights, and are "responsible for outputs and their subsequent uses."[1] There is no user-count or revenue threshold, no list of prohibited applications in the agreement itself, and no requirement that a derivative model's name refer to NVIDIA.[1]
Conditions and termination
Section 2.1 ("Conditions for Use") holds the provisions under which the grant can be lost:[1]
| Provision | What the October 2025 text says |
|---|---|
| Litigation | If the licensee starts copyright or patent litigation against any entity, including a cross-claim or counterclaim, alleging that the model or a derivative model infringes, the licenses for that model end on the date the suit is filed |
| Guardrails | If the licensee bypasses, disables, reduces the efficacy of or circumvents any "Guardrail" (a technical limitation, safety guardrail or guardrail hyperparameter, encryption, security, digital rights management or authentication mechanism) "without a substantially similar Guardrail appropriate for your use case," the rights end automatically |
| Updates | NVIDIA "may update this Agreement to comply with legal and regulatory requirements at any time," and the licensee agrees either to comply with the updated license or to stop copying, using and distributing the model and derivatives |
| Special-purpose models | NVIDIA may mark a model in its documentation as a "Special-Purpose Model," one "only competent in a narrow set of purpose-specific tasks" that should not be used for general-purpose applications |
Section 2.3, headed "AI Ethics," requires that use of the models "must be consistent with NVIDIA's Trustworthy AI terms."[1] Those terms, last modified June 27, 2024, have two parts. The first encourages users to meet the legal and ethical requirements of their industry, to take reasonable measures against unintended bias and harm, including to underrepresented or vulnerable groups, and to inform their own users of a product's nature and limitations. The second "expressly prohibits" use of NVIDIA products or services for any purpose that violates applicable law, giving illegal surveillance, illegal collection or processing of biometric information without required consent, and illegal harassment, abuse, threats or bullying, or intentionally misleading or deceiving others, as examples.[9] Every listed prohibition concerns unlawful use; neither the agreement nor the Trustworthy AI terms bars particular fields such as weapons, medicine or critical infrastructure.[1][9]
Redistribution and attribution
A licensee may reproduce and distribute the model or derivative models, modified or not, in any medium. Whoever distributes the model must give recipients a copy of the agreement and include a "Notice" text file with the line "Licensed by NVIDIA Corporation under the NVIDIA Open Model License."[1] Licensees may put their own copyright and their own license terms on their modifications and on derivative models as a whole, as long as their use of the underlying model still complies with the agreement.[1]
Cosmos models carry an extra condition. Anyone who distributes or makes available an "NVIDIA Cosmos Model" (defined as a multimodal model shared under the agreement), a product or service that contains or uses one, or a derivative of one, or who uses a Cosmos model or its outputs "to create, train, fine tune, or otherwise improve an AI model," must display "Built on NVIDIA Cosmos" on a related website, user interface, blog post, about page or product documentation.[1]
NVIDIA's own product page for Nemotron describes the license as one that "allows users to use, modify, distribute, and commercially deploy the models and derivatives without crediting NVIDIA."[32] In the text, the Notice file obligation is attached to distributing the model, and the "Built on NVIDIA Cosmos" display applies only to Cosmos models.[1]
Other provisions
A "Separate Components" section says that third-party components shipped with a model keep their own licenses, but that the agreement "will prevail" over a third-party open source software license unless that license requires its own terms to prevail.[1] The agreement grants no trademark rights beyond describing the model's origin, disclaims all warranties, limits NVIDIA's liability, and requires the licensee to indemnify NVIDIA against third-party claims arising from use or distribution of the model, derivative models or outputs. It is governed by Delaware and United States law, with exclusive jurisdiction in the state and federal courts of Santa Clara County, California, and requires compliance with United States export and sanctions rules.[1]
Revision history
NVIDIA has not published a changelog. The dates below are the "Last Modified" and "Version Release Date" lines of each text, compared with archived copies of the agreement page.[2][5][6][7][8][1]
| Version date | Main changes |
|---|---|
| June 14, 2024 | First version, released with Nemotron-4 340B. Litigation termination and NVIDIA's right to update; an "AI Ethics" section that encouraged legal and ethical use and prohibited unlawful uses, in nearly the same words as NVIDIA's Trustworthy AI terms of June 27, 2024; no guardrail clause[2][9] |
| January 6, 2025 | Released with Cosmos. Added the guardrail termination clause, replaced the AI Ethics text with a requirement to follow the Trustworthy AI terms, and added the "NVIDIA Cosmos Model" definition and "Built on NVIDIA Cosmos" attribution[5] |
| April 28, 2025 | Added the "Separate Components" section on third-party and open source components[6] |
| June 16, 2025 | Defined "Guardrail" and limited termination to circumvention "without a substantially similar Guardrail appropriate for your use case"[7] |
| September 23, 2025 | Added the "Special-Purpose Model" designation and made licensees responsible for reviewing model documentation and its limitations[8] |
| October 24, 2025 | Replaced the remaining uses of "Model Derivatives" with "Derivative Models"; no change in substance[1] |
Because the agreement lets NVIDIA update it, model cards do not always point to the current text. As of November 2026, 52 of the NVIDIA repositories on Hugging Face that name the license still linked to the June 2024 PDF, among them the Nemotron-4 340B and the original Minitron cards.[28][30]
Nemotron Open Model License and NVIDIA Open Model Agreement
NVIDIA introduced the NVIDIA Nemotron Open Model License, dated December 15, 2025, with the Nemotron 3 family.[11] It keeps the opening statements that works are commercially usable, that derivative works may be created and distributed, and that NVIDIA claims no ownership of outputs, but is otherwise a different document. Its definitions of "Work" and "Derivative Works," and its sections on redistribution and on accepting warranty or additional liability, closely follow the wording of the Apache License 2.0.[11][46] The NVIDIA Open Model Agreement, with a release date of April 2, 2026 and an internal version date of March 9, 2026, uses the same text with NVIDIA's name for the agreement changed.[12]
| Term | NVIDIA Open Model License (Oct. 2025) | Nemotron Open Model License (Dec. 2025) and Open Model Agreement (Apr. 2026) |
|---|---|---|
| Grant | Perpetual, royalty-free, "revocable"; includes make, sell, offer for sale and import | Perpetual, royalty-free, "irrevocable"; reproduce, prepare derivative works, display, perform, sublicense and distribute |
| Litigation trigger | Suit alleging the model or a derivative infringes copyright or patents | Suit alleging the work "or an output from the Work" infringes patents or copyright |
| Guardrail clause | Yes | No |
| Trustworthy AI terms | Required | Not mentioned |
| NVIDIA may update terms | Yes | Not mentioned |
| Attribution | Notice file naming the NVIDIA Open Model License; "Built on NVIDIA Cosmos" for Cosmos models | Keep existing notices; if the work has a NOTICE file, include "Licensed by NVIDIA Corporation under the NVIDIA Nemotron Model License" (or the Open Model Agreement) |
| Indemnity | Third-party claims arising from use or distribution of the model, derivatives or outputs | Same |
| Governing law | Delaware; Santa Clara County courts | Same |
The new license was not applied consistently at first. NVIDIA's technical blog and Hugging Face post for Nemotron 3 Nano, both dated December 15, 2025, said the weights were released under the NVIDIA Open Model License,[19] and the Nano model card linked to that agreement until a December 23, 2025 commit switched it to the Nemotron Open Model License.[20] When Nemotron 3 Super was published on March 11, 2026, its card text named the Nemotron license while its metadata still named the NVIDIA Open Model License; VentureBeat's launch report described the release as being under the NVIDIA Open Model License (updated October 2025), with its guardrail and litigation "red lines."[21] A March 14, 2026 commit changed the metadata to the Nemotron license.[22] The Open Model Agreement was first used for the Nemotron 3 Nano Omni models in April 2026 and later for some Cosmos, robotics and quantized checkpoints.[47][28]
Non-commercial and research licenses
NVIDIA uses several other licenses for checkpoints it does not offer for commercial use.
| License | Date of current text | Main terms | Example models |
|---|---|---|---|
| NVIDIA Internal Scientific Research and Development Model License | April 11, 2025 | Use, reproduction and derivatives only "for the sole purpose of Your internal, scientific research and development and in a non-production environment"; no distribution, deployment or production use; terminates if the licensee brings any legal proceeding against NVIDIA over the models or their outputs[13] | Nemotron-H base and reasoning models, Nemotron-Elastic-12B[28] |
| "NVIDIA License," also published as the NVIDIA One-Way Noncommercial License | March 22, 2022 | Use and redistribution "non-commercially," defined as "for research or evaluation purposes only," while NVIDIA and its affiliates may use the work and derivatives commercially; derivatives must keep the use limitation; ends on a patent claim against the licensor[14] | Isaac GR00T N1 and N1.5 checkpoints[16] |
| NVIDIA Source Code License (project-specific versions) | Varies | Same structure and non-commercial limitation, applied to research code and pretrained networks[15] | StyleGAN3, RADIO vision encoders[15][28] |
| NVIDIA Community Model License | January 30, 2025 | Production use only through NVIDIA NIM under an NVIDIA AI Enterprise subscription, with exceptions for some local RTX uses; no use of models or outputs to improve other AI models without NVIDIA's written approval; no guardrail circumvention[18] | Some ONNX and Riva checkpoints[28] |
The main differences from the Open Model License are commercial use and distribution. The "One-Way" and source code licenses reserve commercial rights to NVIDIA, and the internal research license forbids any distribution or production use. The internal research license, like the Open Model License, says that licensees own their derivative models and that NVIDIA claims no ownership of outputs; the One-Way license does not address outputs. Each of them ends the license on certain lawsuits.[13][14]
The license statements attached to a single release have not always matched. The GR00T N1.7 model card pointed to the NVIDIA Open Model License from the April 2026 release, but the LICENSE file bundled with the checkpoint contained the non-commercial NVIDIA License until a November 10, 2026 commit replaced it with the Open Model License and added the Cosmos attribution.[43] The original Minitron 4B and 8B cards name the Open Model License but also say the model "is for research and development only."[44]
Models released under the license
The table lists families whose weights NVIDIA has published under the NVIDIA Open Model License, according to the license fields of their Hugging Face repositories in November 2026.[28]
| Family | Models under the NVIDIA Open Model License | Other licenses in the family |
|---|---|---|
| Nemotron (language) | Nemotron-4 340B (June 2024); Nemotron Nano 2 (Nemotron-Nano-9B-v2, August 2025); Nemotron-Mini-4B and the Hindi 4B models | Nemotron 3 Nano and Super, Nano-9B-v2-Japanese: Nemotron Open Model License; Nemotron 3 Nano Omni: Open Model Agreement; Nemotron 3 Ultra and Nemotron 3.5: OpenMDW-1.1; Nemotron-H: internal research license[3][19][24][28][47] |
| Llama Nemotron | Llama-3.1-Nemotron-51B, Llama-3.3-Nemotron-Super-49B-v1.5 and others, which also list Meta's Llama license as additional terms and say "Built with Llama"[31] | Llama-3.1-Nemotron-70B Instruct and Reward (2024): Llama 3.1 Community License only |
| Minitron and Hymba | Minitron 4B and 8B, Llama-3.1-Minitron 4B, Mistral-NeMo-Minitron 8B (2024); Hymba-1.5B (October 2024) | |
| NVIDIA Cosmos | Cosmos 1.0 (January 2025), Predict1, Transfer1, Predict2, Cosmos Reason 1 and 2, Predict2.5 and Transfer2.5 | Cosmos 3 (June 2026): OpenMDW-1.1[25] |
| Isaac GR00T | GR00T N1.7 (April 2026) and GR00T-H-N1.7 checkpoints | N1, N1.5 and N1.6, and the first GR00T-H release: non-commercial NVIDIA License[16][43] |
| NVIDIA Ising | Ising-Calibration-1 and Ising-Decoder-SurfaceCode-1 (March-April 2026) | Ising-Calibration-1.5 and later decoders (July 2026): OpenMDW-1.1 |
| NVIDIA Earth-2 | Atlas, StormScope GOES-MRMS, HealDA | FourCastNet 3 and cBottle: Apache 2.0; HealDA v2 and later StormScope models: OpenMDW-1.1[41] |
| Speech | Parakeet multitalker streaming, end-of-utterance, Vietnamese, Danish and unified models (2025-2026); Nemotron Speech streaming ASR | The general-purpose Parakeet CTC, RNNT and TDT checkpoints (including parakeet-tdt-0.6b-v3, August 2025) and the Canary checkpoints: CC BY 4.0 (Canary-1B: CC BY-NC 4.0); Nemotron 3.5 ASR: OpenMDW-1.1[42] |
| Other | C-RADIO vision encoders, BioNeMo biology and chemistry models, Audio2Face, and NVFP4 and speculative-decoding conversions of third-party models such as Kimi and Qwen | DrivAerML CAE models and Kimi-K3-NVFP4: metadata names the Open Model License, card text the Open Model Agreement |
Synthetic data generation was the use NVIDIA emphasized in 2024.[4] Because an output is not a derivative model and NVIDIA claims no ownership of outputs, the agreement attaches no redistribution or attribution conditions to generated data; for Cosmos models, however, using a model's outputs to train or improve another AI model triggers the "Built on NVIDIA Cosmos" requirement.[1]
Move to OpenMDW
On May 28, 2026 the Linux Foundation released OpenMDW-1.1 and announced that NVIDIA "plans to adopt the license framework across future releases of its Cosmos, Isaac GR00T, Ising, and Nemotron open model families." The foundation described OpenMDW as an alternative to "custom AI licenses that imposed restrictive usage limitations."[23] Kari Briski, NVIDIA's vice president of generative AI, said: "By adopting the Linux Foundation's OpenMDW framework for NVIDIA open model families, we're helping establish a simpler, more consistent standard for open models at scale."[23] OpenMDW has no guardrail, ethics or update clause; its conditions are keeping the license and notices with redistributed copies, and rights end for anyone who sues claiming the materials infringe a patent or copyright.[49]
The first large releases followed within days. Cosmos 3 was published under OpenMDW-1.1 in June 2026,[25] and NVIDIA's June 4, 2026 post on Nemotron 3 Ultra said "Nemotron model releases are moving to OpenMDW-1.1."[24] Later OpenMDW releases included NVIDIA Alpamayo 2 Super (August 4, 2026), Nemotron 3.5 Lightning (August 2026), NVIDIA Ising checkpoints and HealDA v2.[28][41][48] NVIDIA also relicensed some earlier models: on August 26, 2026 it changed the llama-nemotron-rerank-1b-v2 retrieval model from the NVIDIA Open Model License to OpenMDW-1.1,[26] and on August 4, 2026 it moved the Alpamayo-R1-10B weights from a non-commercial license to OpenMDW-1.1.[27]
The change was not complete by November 2026. Repositories created in October and November 2026 still named the NVIDIA Open Model License, including Cosmos3-H-Dreams and other Cosmos 3 derived checkpoints, GR00T N1.7 fine-tunes, and quantized third-party models; the Nemotron-Labs Sigma research models used the Nemotron Open Model License, and the NVFP4 conversion of Kimi K3.1 Pro used the Open Model Agreement.[28][45] On November 17, 2026 the NVIDIA account held 1,059 model repositories: 334 named the NVIDIA Open Model License (330 as nvidia-open-model-license and 4 under a variant name), 103 OpenMDW-1.1, 34 the Nemotron Open Model License, 20 the Open Model Agreement and 14 the internal research license, counting variant spellings of each name.[28] The count follows the license_name field. For at least seven of these repositories the card's own text names a different license: the Open Model Agreement for the DrivAerML CAE models and Kimi-K3-NVFP4, and the non-commercial NVIDIA License for MetricTrail.[28]
Status as an open source license
The NVIDIA Open Model License is not among the licenses approved by the Open Source Initiative, and the SPDX License List, as of its version 3.29.0 (September 16, 2026), has no identifier for any NVIDIA model license.[50][35] Published positions differ:
- The European Open Source AI Index, an academic project based in the Netherlands that rates the openness of generative models, records the license of Llama-3.3-Nemotron-Super-49B-v1.5 as "NVIDIA Open Model License Agreement, not an OSI-approved license."[36]
- The ScanCode LicenseDB, a reference database used by license-scanning tools, classifies the NVIDIA Open Model License Agreement 2025 as "Proprietary Free" and the Nemotron Open Model License as "Permissive."[33][34]
- Artificial Analysis's Openness Index (version 1.0) uses Nemotron Nano 9B v2, an NVIDIA Open Model License release, as its example of a weights license that allows commercial use without attribution "but may include other limitations," the second-highest of four grades; the top grade is for licenses such as MIT and Apache 2.0.[37]
- VentureBeat's Carl Franzen wrote in March 2026 that the license "provides a permissive framework for enterprise adoption" but carries "safeguard" clauses "that differentiate it from pure open-source licenses like MIT or Apache 2.0."[21]
- NVIDIA's own Nemotron page says the models "aren't just open, but truly open source," because NVIDIA also publishes training datasets and techniques.[32] Reviewing Nemotron 3 Nano for heise in December 2025, Christian Winkler called the release, with its open weights and published training data, "an almost open-source model."[38]
Under the OSI's Open Source AI Definition, a license on the weights does not by itself settle the question, since the definition also requires the training code and detailed information about the training data under open terms.[40] OpenMDW-1.1, the license NVIDIA moved to, was submitted to the OSI's license-review process in August 2026.[39]
Identification in model hubs
Hugging Face's list of predefined licenses does not include any NVIDIA license, so NVIDIA repositories set license: other in their model card metadata and name the license in the free-text license_name field, with a license_link to the text.[29] The usual name is nvidia-open-model-license, linked either to the nvidia.com page or to the June 2024 PDF; other repositories use nvidia-open-model-license-agreement, nvidia-nemotron-open-model-license, nvidia-open-model-agreement or nvidia-internal-scientific-research-and-development-model-license.[28][30] Because these are free-text values, Hugging Face's license filter does not list them: a search for the license tag nvidia-open-model-license returned no models on November 17, 2026, and NVIDIA's models under the agreement appear only under the general "other" tag.[51][28] NVIDIA's OpenMDW repositories are split the same way: some use the Hub's predefined openmdw-1.1 license, which the filter finds, and others use license: other with openmdw-1.1 or openmdw1.1-license as the name.[28][29]
Continue reading
- OpenMDWOpenMDW (Open Model, Data and Weights License) is a permissive license agreement developed under the Linux Foundation specifically for machine learning models and their related artifacts.Mentioned 21 times above
- OpenRAILOpenRAIL (Open Responsible AI Licenses, also written Open RAIL) is a family of licenses for machine learning models, source code, data and applications that gives royalty-free permission to use, modify and…Listed in See also
- Open-Weight LLM License ComparisonAs of November 2026, the most widely used licenses that put no user, revenue or use conditions on commercial use of open-weight LLMs are two standard, OSI-approved ones: Apache-2.0 (Alibaba's Qwen3, most of…Listed in See also
- List of open-weight modelsThis is a list of open-weight models: artificial intelligence models whose trained parameters (weights) the developer has published for download, so that the model can be run, studied and in most cases…Listed in See also
- Model Openness FrameworkThe Model Openness Framework (MOF) is a classification system from the Linux Foundation's LF AI & Data Foundation that grades a released machine learning model on two things: how many of the artifacts from its…Listed in See also
- OpenwashingOpenwashing (also written open-washing or open washing) is the practice of presenting a product, organisation or technology as "open" while withholding the things that openness normally implies.Listed in See also
More to read
References
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20NVIDIA. "NVIDIA Open Model License Agreement" (Last Modified: October 24, 2025). NVIDIA, accessed November 17, 2026. nvidia.com/...nvidia-open-model-license
- ^1 ^2 ^3NVIDIA. "NVIDIA Open Model License Agreement" (Version Release Date: June 14, 2024). NVIDIA Developer downloads. developer.download.nvidia.com/...ent-june-2024.pdf
- ^1 ^2 ^3NVIDIA Research. "Nemotron-4 340B." June 14, 2024. research.nvidia.com/...2024-06_nemotron-4-340b
- ^1 ^2Ankit Patel. "NVIDIA Releases Open Synthetic Data Generation Pipeline for Training Large Language Models." NVIDIA Blog, June 14, 2024. blogs.nvidia.com/...c-data-generation-llm-training
- ^1 ^2 ^3 ^4NVIDIA. "NVIDIA Open Model License Agreement" (Last Modified: January 6, 2025). Internet Archive copy of January 12, 2025. web.archive.org/...nvidia-open-model-license
- ^1 ^2 ^3NVIDIA. "NVIDIA Open Model License Agreement" (Last Modified: April 28, 2025). Internet Archive copy of May 20, 2025. web.archive.org/...nvidia-open-model-license
- ^1 ^2 ^3NVIDIA. "NVIDIA Open Model License Agreement" (Last Modified: June 16, 2025). Internet Archive copy of July 26, 2025. web.archive.org/...nvidia-open-model-license
- ^1 ^2 ^3NVIDIA. "NVIDIA Open Model License Agreement" (Last Modified: September 23, 2025). Internet Archive copy of October 1, 2025. web.archive.org/...nvidia-open-model-license
- ^1 ^2 ^3NVIDIA. "Trustworthy AI Terms" (Last Modified: June 27, 2024). NVIDIA, accessed November 17, 2026. nvidia.com/...terms
- ^Ming-Yu Liu. "NVIDIA Makes Cosmos World Foundation Models Openly Available to Physical AI Developer Community." NVIDIA Blog, January 6, 2025. blogs.nvidia.com/...cosmos-world-foundation-models
- ^1 ^2 ^3NVIDIA. "NVIDIA Nemotron Open Model License" (Last Modified: December 15, 2025). NVIDIA, accessed November 17, 2026. nvidia.com/...nvidia-nemotron-open-model-license
- ^1 ^2NVIDIA. "NVIDIA Open Model Agreement" (Release Date: April 2, 2026; v. March 9, 2026). NVIDIA, accessed November 17, 2026. nvidia.com/...nvidia-open-model-agreement
- ^1 ^2 ^3NVIDIA. "NVIDIA Internal Scientific Research and Development Model License" (Last Modified: April 11, 2025). NVIDIA, accessed November 17, 2026. nvidia.com/...search-and-development-model-license
- ^1 ^2 ^3NVIDIA. "NVIDIA License" (NVIDIA One-Way Noncommercial License, March 22, 2022). NVIDIA Developer downloads. developer.download.nvidia.com/...nse-22Mar2022.pdf
- ^1 ^2NVIDIA. "NVIDIA Source Code License for StyleGAN3." GitHub, NVlabs/stylegan3, 2021. github.com/...LICENSE.txt
- ^1 ^2NVIDIA. "GR00T-N1-2B: LICENSE" and "GR00T-N1.5-3B" (model card linking the NVIDIA One-Way Noncommercial License). Hugging Face. huggingface.co/...LICENSE ; huggingface.co/...GR00T-N1.5-3B
- ^1 ^2NVIDIA. "NVIDIA AI Foundation Models Community License Agreement" (November 10, 2023). NVIDIA Developer downloads. developer.download.nvidia.com/...nse-10Nov2023.pdf
- ^NVIDIA. "NVIDIA Community Model License" (Last Modified: January 30, 2025). NVIDIA, accessed November 17, 2026. nvidia.com/...nvidia-community-models-license
- ^1 ^2Chris Alexiuk, Shashank Verma and Chintan Patel. "Inside NVIDIA Nemotron 3: Techniques, Tools, and Data That Make It Efficient and Accurate." NVIDIA Technical Blog, December 15, 2025; and NVIDIA, "Nemotron 3 Nano: A New Standard for Efficient, Open, and Intelligent Agentic Models," Hugging Face blog, December 15, 2025. developer.nvidia.com/...-it-efficient-and-accurate ; huggingface.co/...fficient-open-intelligent-models
- ^NVIDIA. "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16: Update License (#32)." Hugging Face commit ad735b6, December 23, 2025. huggingface.co/...4a5e53625fe46dedae74b3f72a1c6ed7
- ^1 ^2Carl Franzen. "Nvidia's new open weights Nemotron 3 super combines three different architectures to beat gpt-oss and Qwen in throughput." VentureBeat, March 11, 2026. venturebeat.com/...-super-combines-three-different venturebeat.com/...-super-combines-three-different
- ^NVIDIA. "NVIDIA-Nemotron-3-Super-120B-A12B-BF16: Update License." Hugging Face commit 49ad1f4, March 14, 2026. huggingface.co/...49ad1f46ee
- ^1 ^2 ^3The Linux Foundation. "Linux Foundation Releases OpenMDW-1.1; NVIDIA Adopts OpenMDW for Cosmos, Isaac GR00T, Ising and Nemotron AI Model Families." May 28, 2026. linuxfoundation.org/...-nemotron-ai-model-families
- ^1 ^2Chris Alexiuk and Chintan Patel. "NVIDIA Nemotron 3 Ultra Powers Faster, More Efficient Reasoning for Long-Running Agents." NVIDIA Technical Blog, June 4, 2026. developer.nvidia.com/...ng-for-long-running-agents
- ^1 ^2NVIDIA. "Cosmos3-Nano" (model card; license: OpenMDW 1.1). Hugging Face, accessed November 17, 2026. huggingface.co/...Cosmos3-Nano
- ^NVIDIA. "llama-nemotron-rerank-1b-v2: Update license to OpenMDW-1.1 (#16)." Hugging Face commit, August 26, 2026. huggingface.co/...3160b397bfd5
- ^NVIDIA. "Alpamayo-R1-10B: Update README.md" (weights license changed to OpenMDW-1.1). Hugging Face commit, August 4, 2026. huggingface.co/...c69ab9a2a31a
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17Hugging Face. "Models by nvidia" (Hub API listing with model card metadata; 1,059 repositories retrieved November 17, 2026). huggingface.co/...models
- ^1 ^2Hugging Face. "Licenses" (Hub documentation). Accessed November 17, 2026. huggingface.co/...repositories-licenses
- ^1 ^2 ^3NVIDIA. "Nemotron-4-340B-Instruct" (model card; license_link to the June 2024 agreement). Hugging Face. huggingface.co/...Nemotron-4-340B-Instruct
- ^NVIDIA. "Llama-3_3-Nemotron-Super-49B-v1_5" (model card, governing terms). Hugging Face. huggingface.co/...Llama-3_3-Nemotron-Super-49B-v1_5
- ^1 ^2NVIDIA. "NVIDIA Nemotron" (product page and FAQ). Accessed November 17, 2026. nvidia.com/...nemotron
- ^AboutCode. "NVIDIA Open Model License Agreement 2025" (key nvidia-open-model-2025). ScanCode LicenseDB. scancode-licensedb.aboutcode.org/...pen-model-2025
- ^AboutCode. "NVIDIA Nemotron Open Model License 2025-12-15" (key nemotron-open-model-2025-12-15). ScanCode LicenseDB. scancode-licensedb.aboutcode.org/...del-2025-12-15
- ^1 ^2SPDX Workgroup, Linux Foundation. "SPDX License List" (version 3.29.0, September 16, 2026). spdx.org/licenses
- ^Open Source AI Index. "Llama Nemotron" (model entry; index generated August 27, 2026). osai-index.eu/...llama-nemotron
- ^Artificial Analysis. "Artificial Analysis Openness Index: Specification V1.0." 2025. artificialanalysis.ai/...nness-Index-Spec_V1-0.pdf
- ^Christian Winkler. "Innovative and almost completely open-source: Nvidia Nemotron 3 Nano." heise online, December 19, 2025. heise.de/-11120837
- ^Michael Dolan. "For Approval: OpenMDW License Agreement, versions 1.1 (OpenMDW-1.1)." OSI license-review mailing list, August 2026. lists.opensource.org/...006125
- ^Open Source Initiative. "The Open Source AI Definition 1.0." opensource.org/...open-source-ai-definition
- ^1 ^2NVIDIA. Earth-2 checkpoint model cards: "atlas-era5" and "healda" (NVIDIA Open Model License), "healda-v2" (OpenMDW-1.1), "fourcastnet3" and "cbottle" (Apache 2.0). Hugging Face, accessed November 17, 2026. huggingface.co/...atlas-era5 ; huggingface.co/...healda ; huggingface.co/...healda-v2
- ^NVIDIA. Speech model cards: "parakeet-unified-en-0.6b" and "nemotron-speech-streaming-en-0.6b" (NVIDIA Open Model License), "parakeet-tdt-0.6b-v3" and "canary-1b-v2" (CC BY 4.0), "canary-1b" (CC BY-NC 4.0). Hugging Face, accessed November 17, 2026. huggingface.co/...parakeet-unified-en-0.6b ; huggingface.co/...canary-1b-v2
- ^1 ^2NVIDIA. "GR00T-N1.7-3B" (model card) and commit 373d34e, "Use the NVIDIA Open Model License and add the Cosmos attribution (#5)," November 10, 2026. Hugging Face. huggingface.co/...main
- ^NVIDIA. "Minitron-8B-Base" (model card). Hugging Face. huggingface.co/...Minitron-8B-Base
- ^NVIDIA. "Cosmos3-H-Dreams" (model card; license: NVIDIA Open Model License). Hugging Face, accessed November 17, 2026. huggingface.co/...Cosmos3-H-Dreams
- ^Apache Software Foundation. "Apache License, Version 2.0." January 2004. apache.org/...LICENSE-2.0
- ^1 ^2NVIDIA. "Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16" (model card; governing terms: NVIDIA Open Model Agreement). Hugging Face. huggingface.co/...Nano-Omni-30B-A3B-Reasoning-BF16
- ^NVIDIA. "Alpamayo2-Super" (model card; license: OpenMDW-1.1). Hugging Face. huggingface.co/...Alpamayo2-Super
- ^OpenMDW project. "OpenMDW License Agreement, version 1.1 (OpenMDW-1.1)." GitHub. github.com/...LICENSE.OpenMDW-1.1
- ^1 ^2Open Source Initiative. "OSI Approved Licenses." Accessed November 17, 2026. opensource.org/licenses
- ^Hugging Face. "Models with the license tag nvidia-open-model-license" (Hub API query; no results on November 17, 2026). huggingface.co/...models
Improve this article
Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.
1 revision · v1 · 4,672 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Reviewer note: Independent verifier checked 118 claims against 74 sources incl. archived license revisions and a recount of 1,059 NVIDIA Hugging Face repos; 12 minor fixes applied.
Cite this page: AI Wiki. "NVIDIA Open Model License." aiwiki.ai, updated 17 Nov 2026, fact-checked 17 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/nvidia_open_model_license