Citation and evidence

NVIDIA Open Model License

23 min full readUpdated 51 references

This article's verification

Report a problem with this article

Edit article
More

Use this article

Raw MarkdownExplore connections

Improve this page

Revision historyDiscuss this article

Topics

AI Policy & RegulationOpen Source AI

Cite this article

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]

ProvisionWhat the October 2025 text says
LitigationIf 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
GuardrailsIf 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
UpdatesNVIDIA "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 modelsNVIDIA 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

Expanded article table

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 dateMain changes
June 14, 2024First 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, 2025Released 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, 2025Added the "Separate Components" section on third-party and open source components[6]
June 16, 2025Defined "Guardrail" and limited termination to circumvention "without a substantially similar Guardrail appropriate for your use case"[7]
September 23, 2025Added the "Special-Purpose Model" designation and made licensees responsible for reviewing model documentation and its limitations[8]
October 24, 2025Replaced the remaining uses of "Model Derivatives" with "Derivative Models"; no change in substance[1]

Expanded article table

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]

TermNVIDIA Open Model License (Oct. 2025)Nemotron Open Model License (Dec. 2025) and Open Model Agreement (Apr. 2026)
GrantPerpetual, royalty-free, "revocable"; includes make, sell, offer for sale and importPerpetual, royalty-free, "irrevocable"; reproduce, prepare derivative works, display, perform, sublicense and distribute
Litigation triggerSuit alleging the model or a derivative infringes copyright or patentsSuit alleging the work "or an output from the Work" infringes patents or copyright
Guardrail clauseYesNo
Trustworthy AI termsRequiredNot mentioned
NVIDIA may update termsYesNot mentioned
AttributionNotice file naming the NVIDIA Open Model License; "Built on NVIDIA Cosmos" for Cosmos modelsKeep 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)
IndemnityThird-party claims arising from use or distribution of the model, derivatives or outputsSame
Governing lawDelaware; Santa Clara County courtsSame

Expanded article table

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.

LicenseDate of current textMain termsExample models
NVIDIA Internal Scientific Research and Development Model LicenseApril 11, 2025Use, 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 LicenseMarch 22, 2022Use 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)VariesSame structure and non-commercial limitation, applied to research code and pretrained networks[15]StyleGAN3, RADIO vision encoders[15][28]
NVIDIA Community Model LicenseJanuary 30, 2025Production 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]

Expanded article table

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]

FamilyModels under the NVIDIA Open Model LicenseOther 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 modelsNemotron 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 NemotronLlama-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 HymbaMinitron 4B and 8B, Llama-3.1-Minitron 4B, Mistral-NeMo-Minitron 8B (2024); Hymba-1.5B (October 2024)
NVIDIA CosmosCosmos 1.0 (January 2025), Predict1, Transfer1, Predict2, Cosmos Reason 1 and 2, Predict2.5 and Transfer2.5Cosmos 3 (June 2026): OpenMDW-1.1[25]
Isaac GR00TGR00T N1.7 (April 2026) and GR00T-H-N1.7 checkpointsN1, N1.5 and N1.6, and the first GR00T-H release: non-commercial NVIDIA License[16][43]
NVIDIA IsingIsing-Calibration-1 and Ising-Decoder-SurfaceCode-1 (March-April 2026)Ising-Calibration-1.5 and later decoders (July 2026): OpenMDW-1.1
NVIDIA Earth-2Atlas, StormScope GOES-MRMS, HealDAFourCastNet 3 and cBottle: Apache 2.0; HealDA v2 and later StormScope models: OpenMDW-1.1[41]
SpeechParakeet multitalker streaming, end-of-utterance, Vietnamese, Danish and unified models (2025-2026); Nemotron Speech streaming ASRThe 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]
OtherC-RADIO vision encoders, BioNeMo biology and chemistry models, Audio2Face, and NVFP4 and speculative-decoding conversions of third-party models such as Kimi and QwenDrivAerML CAE models and Kimi-K3-NVFP4: metadata names the Open Model License, card text the Open Model Agreement

Expanded article table

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

More to read

Topics:AI Policy & RegulationOpen Source AI

References

  1. ^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
  2. ^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
  3. ^1 ^2 ^3NVIDIA Research. "Nemotron-4 340B." June 14, 2024. research.nvidia.com/...2024-06_nemotron-4-340b
  4. ^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
  5. ^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
  6. ^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
  7. ^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
  8. ^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
  9. ^1 ^2 ^3NVIDIA. "Trustworthy AI Terms" (Last Modified: June 27, 2024). NVIDIA, accessed November 17, 2026. nvidia.com/...terms
  10. ^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
  11. ^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
  12. ^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
  13. ^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
  14. ^1 ^2 ^3NVIDIA. "NVIDIA License" (NVIDIA One-Way Noncommercial License, March 22, 2022). NVIDIA Developer downloads. developer.download.nvidia.com/...nse-22Mar2022.pdf
  15. ^1 ^2NVIDIA. "NVIDIA Source Code License for StyleGAN3." GitHub, NVlabs/stylegan3, 2021. github.com/...LICENSE.txt
  16. ^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
  17. ^1 ^2NVIDIA. "NVIDIA AI Foundation Models Community License Agreement" (November 10, 2023). NVIDIA Developer downloads. developer.download.nvidia.com/...nse-10Nov2023.pdf
  18. ^NVIDIA. "NVIDIA Community Model License" (Last Modified: January 30, 2025). NVIDIA, accessed November 17, 2026. nvidia.com/...nvidia-community-models-license
  19. ^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
  20. ^NVIDIA. "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16: Update License (#32)." Hugging Face commit ad735b6, December 23, 2025. huggingface.co/...4a5e53625fe46dedae74b3f72a1c6ed7
  21. ^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
  22. ^NVIDIA. "NVIDIA-Nemotron-3-Super-120B-A12B-BF16: Update License." Hugging Face commit 49ad1f4, March 14, 2026. huggingface.co/...49ad1f46ee
  23. ^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
  24. ^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
  25. ^1 ^2NVIDIA. "Cosmos3-Nano" (model card; license: OpenMDW 1.1). Hugging Face, accessed November 17, 2026. huggingface.co/...Cosmos3-Nano
  26. ^NVIDIA. "llama-nemotron-rerank-1b-v2: Update license to OpenMDW-1.1 (#16)." Hugging Face commit, August 26, 2026. huggingface.co/...3160b397bfd5
  27. ^NVIDIA. "Alpamayo-R1-10B: Update README.md" (weights license changed to OpenMDW-1.1). Hugging Face commit, August 4, 2026. huggingface.co/...c69ab9a2a31a
  28. ^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
  29. ^1 ^2Hugging Face. "Licenses" (Hub documentation). Accessed November 17, 2026. huggingface.co/...repositories-licenses
  30. ^1 ^2 ^3NVIDIA. "Nemotron-4-340B-Instruct" (model card; license_link to the June 2024 agreement). Hugging Face. huggingface.co/...Nemotron-4-340B-Instruct
  31. ^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
  32. ^1 ^2NVIDIA. "NVIDIA Nemotron" (product page and FAQ). Accessed November 17, 2026. nvidia.com/...nemotron
  33. ^AboutCode. "NVIDIA Open Model License Agreement 2025" (key nvidia-open-model-2025). ScanCode LicenseDB. scancode-licensedb.aboutcode.org/...pen-model-2025
  34. ^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
  35. ^1 ^2SPDX Workgroup, Linux Foundation. "SPDX License List" (version 3.29.0, September 16, 2026). spdx.org/licenses
  36. ^Open Source AI Index. "Llama Nemotron" (model entry; index generated August 27, 2026). osai-index.eu/...llama-nemotron
  37. ^Artificial Analysis. "Artificial Analysis Openness Index: Specification V1.0." 2025. artificialanalysis.ai/...nness-Index-Spec_V1-0.pdf
  38. ^Christian Winkler. "Innovative and almost completely open-source: Nvidia Nemotron 3 Nano." heise online, December 19, 2025. heise.de/-11120837
  39. ^Michael Dolan. "For Approval: OpenMDW License Agreement, versions 1.1 (OpenMDW-1.1)." OSI license-review mailing list, August 2026. lists.opensource.org/...006125
  40. ^Open Source Initiative. "The Open Source AI Definition 1.0." opensource.org/...open-source-ai-definition
  41. ^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
  42. ^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
  43. ^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
  44. ^NVIDIA. "Minitron-8B-Base" (model card). Hugging Face. huggingface.co/...Minitron-8B-Base
  45. ^NVIDIA. "Cosmos3-H-Dreams" (model card; license: NVIDIA Open Model License). Hugging Face, accessed November 17, 2026. huggingface.co/...Cosmos3-H-Dreams
  46. ^Apache Software Foundation. "Apache License, Version 2.0." January 2004. apache.org/...LICENSE-2.0
  47. ^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
  48. ^NVIDIA. "Alpamayo2-Super" (model card; license: OpenMDW-1.1). Hugging Face. huggingface.co/...Alpamayo2-Super
  49. ^OpenMDW project. "OpenMDW License Agreement, version 1.1 (OpenMDW-1.1)." GitHub. github.com/...LICENSE.OpenMDW-1.1
  50. ^1 ^2Open Source Initiative. "OSI Approved Licenses." Accessed November 17, 2026. opensource.org/licenses
  51. ^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

Edit article