Llama Community License
The Llama Community License is the family of license agreements under which Meta has distributed the weights, code and documentation of its Llama large language models since July 2023. Each generation has its own agreement: the Llama 2 Community License Agreement, the Meta Llama 3 Community License Agreement, and the Llama 3.1, 3.2, 3.3 and Llama 4 Community License Agreements, each paired with a separate Acceptable Use Policy that the agreement incorporates by reference.[1][2][3][4][5][6] They were preceded by the LLaMA License Agreement of February 2023, which allowed only non-commercial research use.[13][14]
The community licenses grant a free, worldwide right to use, copy, modify and redistribute the models, including commercially, but with conditions that standard software licenses do not have: companies whose products had more than 700 million monthly active users when a version was released must ask Meta for a separate license, distributors must carry Meta's attribution and naming requirements, and every user must follow the Acceptable Use Policy.[1][6] From Llama 3.2 onward the policy also withholds the license for multimodal models from individuals and companies based in the European Union.[10][12] Meta has repeatedly called Llama "open source"; the Open Source Initiative (OSI) and the Free Software Foundation (FSF) have said the licenses are neither open source nor free software, and AI Wiki describes the models as open weights.[20][21][22] As of November 2026, Meta's newer open-weight models, released under the Muse name, use different terms.[33][34]
History
LLaMA research license (2023)
Meta announced LLaMA on February 24, 2023. Its announcement said: "To maintain integrity and prevent misuse, we are releasing our model under a noncommercial license focused on research use cases", with access granted case by case to academic, government, civil-society and industry research applicants.[13] The inference code in the public repository was released under the GNU General Public License version 3, while the weights were sent to approved applicants who filled in a request form.[15]
The LLaMA License Agreement granted a "non-exclusive, worldwide, non-transferable, non-sublicensable, revocable, royalty free and limited license" to reproduce, distribute and create derivative works of the software "solely for your non-commercial research purposes".[14] It barred any commercial or production use, military and nuclear uses, surveillance, and biometric processing, and required redistributed copies to carry a notice that "LLaMA is licensed under the LLaMA License".[14] On March 3, 2023, about a week after access requests opened, a torrent of the weights was posted on 4chan and spread to other sites.[16] Early fine-tunes such as Stanford's Alpaca were likewise released for research use only.[37]
Llama 2 Community License (July 2023)
With Llama 2, announced on July 18, 2023 together with Microsoft, Meta changed the terms. It described the model as available "free of charge for research and commercial use".[17] The new Llama 2 Community License Agreement introduced the structure that later versions kept: a broad royalty-free grant, the 700 million monthly active user clause, an incorporated Acceptable Use Policy, and a clause that barred using Llama 2 or its outputs to improve any other large language model.[1][7]
Llama 3 and Llama 3.x (2024)
The Meta Llama 3 Community License (April 18, 2024) kept the Llama 2 terms and added branding: distributors had to display "Built with Meta Llama 3", and a model created or improved with the Llama materials had to start its name with "Llama 3".[2] The Llama 3.1 agreement (July 23, 2024) shortened these requirements to "Built with Llama" and a "Llama" name prefix, and removed the ban on improving other models.[3] Meta's announcement said it had "made changes to our license, allowing developers to use the outputs from Llama models" to improve other models, including the 405B model.[18] The Llama 3.2 (September 25, 2024) and Llama 3.3 (December 6, 2024) agreements are the Llama 3.1 text with the version name, dates and links updated; the changes in these releases were made to the Acceptable Use Policy instead.[4][5][10][11]
Llama 4 (2025)
The Llama 4 Community License Agreement carries a "Llama 4 Version Effective Date" of April 5, 2025, the day Meta announced Llama 4 Scout and Maverick.[6][25] Apart from the version name, date and links, its text matches the Llama 3.1 to 3.3 agreements.[3][6]
Terms of the community licenses
Grant and contracting parties
Every community license grants "a non-exclusive, worldwide, non-transferable and royalty-free limited license under Meta's intellectual property or other rights owned by Meta embodied in the Llama Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Llama Materials".[1][6] "Llama Materials" covers the model weights, the inference, training and fine-tuning code, and the documentation. The contracting party is Meta Platforms Ireland Limited for licensees located in, or with their principal place of business in, the European Economic Area or Switzerland, and Meta Platforms, Inc. for everyone else.[1][6]
Acceptance is by clicking "I Accept" or "by using or distributing any portion or element of the Llama Materials".[1] Other recurring terms:
| Clause | What it says |
|---|---|
| Ownership of derivatives | As between the licensee and Meta, the licensee owns the derivative works and modifications it makes, subject to Meta's ownership of the Llama Materials.[1][6] |
| Patent and IP claims; indemnity | A licensee that sues Meta or anyone else claiming that the Llama Materials or their outputs infringe its intellectual property loses its license from the date of filing. Separately, every licensee must indemnify Meta against third-party claims arising from its use or distribution of the Llama Materials.[1][6] |
| Termination | Meta may terminate the agreement for breach; on termination the licensee must delete and stop using the materials.[1][6] |
| Trademarks | No trademark license is granted, except (from Llama 3 onward) a license to use "Llama" or "Llama 3" as needed to meet the naming rule, under Meta's brand guidelines.[2][3] |
| Warranty and liability | The materials and their outputs are provided "as is", and Meta excludes liability for lost profits and indirect damages.[1][6] |
| Governing law | California law, with exclusive jurisdiction in California courts.[1][6] |
The 700 million user clause
Section 2 of each agreement, headed "Additional Commercial Terms", reads in the Llama 2 version: "If, on the Llama 2 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee's affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion".[1] Until Meta grants that license, the company may not exercise any rights under the agreement. Later versions repeat the wording with their own version names.[2][3][6]
As written, the test has four parts. It counts all products or services made available by or for the licensee and its affiliates, not only products that use Llama. It looks at a single month, the calendar month before the version's release date. It is applied once per version, so each new Llama generation sets a new reference date. And it compares the count against a fixed 700 million.[1][6] The agreements contain no clause for a licensee that crosses 700 million after a version's release date. The Llama 4 agreement refers to "the Llama 4 version release date" in section 2 but gives an "Effective Date" of April 5, 2025 in its header.[6] Section 1(b)(ii) exempts from section 2 anyone who receives the Llama Materials as part of an integrated end-user product from a licensee, so the users of a product built on Llama are not tested.[1][6]
The OSI said in 2023 that it did not question "Meta's desire to limit the use of Llama for competitive purposes", but that such a limit takes the license out of the open-source category.[20]
Attribution, "Built with Llama" and naming
Under Llama 2, a distributor had to give recipients a copy of the agreement and keep a "Notice" text file stating that "Llama 2 is licensed under the LLAMA 2 Community License, Copyright (c) Meta Platforms, Inc. All Rights Reserved."[1] There was no branding requirement.
Llama 3 added two obligations for anyone who distributes the materials, a derivative, or "a product or service that uses any of them, including another AI model": prominently display "Built with Meta Llama 3" on a related website, user interface, blog post, about page or product documentation, and, for an AI model created, trained, fine-tuned or improved with the Llama materials and then distributed or made available, include "Llama 3" at the beginning of the model's name.[2] From Llama 3.1 the display text became "Built with Llama", the obligation covered a product or service "that contains" the materials rather than one that "uses" them, the required prefix became "Llama", and the naming rule was extended to models built with "any outputs or results of the Llama Materials".[3][6] Derivative names follow the rule unevenly: Ai2's Tülu 3 checkpoints (published as Llama-3.1-Tulu-3) and NVIDIA's Llama-3.1-Nemotron-70B-Instruct start with "Llama", while Nous Research's Hermes-3-Llama-3.1 models place it later in the name.[28][29][30]
Using outputs to improve other models
The Llama 2 agreement said: "You will not use the Llama Materials or any output or results of the Llama Materials to improve any other large language model (excluding Llama 2 or derivative works thereof)."[1] The Llama 3 agreement kept the same clause with "Meta Llama 3" substituted.[2] In effect it barred using Llama outputs as synthetic data, or for knowledge distillation, to improve another large language model that was not itself a Llama derivative.
The Llama 3.1 agreement removed the clause.[3] Use of outputs to train another model is now allowed, but the model then falls under the naming rule if it is distributed or made available, for example through an API. An April 2025 letter from members of the European Parliament still described Meta as prohibiting "the use of its Llama models for the purpose of training other AI systems"; that description matches the Llama 2 and Llama 3 agreements but not the 3.1 and later texts.[26][3]
Acceptable Use Policy
Prohibited uses
Each agreement makes compliance with a version-specific Acceptable Use Policy a condition of use, and the policy is "hereby incorporated by reference".[1][6] The Llama 2 to 3.1 policies are organized under four headings, and the Llama 3.2 and later policies add a fifth (see below). Users agree not to use, or allow others to use, the model to:[7][12]
| Category | Examples listed in the policy |
|---|---|
| Violate the law or others' rights | Violence or terrorism; exploitation of children or failure to report child sexual abuse material; human trafficking; illegal distribution of material to minors; harassment; unlawful discrimination in employment, credit, housing and other essential services; unlicensed practice of professions such as law, medicine or finance; collecting or inferring sensitive personal information without legal right; infringing third-party rights; creating malware |
| Risk of death or bodily harm | Military, warfare, nuclear and espionage uses and activities subject to the US International Traffic in Arms Regulations; guns and illegal weapons; illegal drugs; operating critical infrastructure, transportation or heavy machinery; self-harm; content inciting violence |
| Deceiving or misleading others | Fraud and disinformation; defamation; spam; impersonation; representing model outputs as human-generated; fake reviews and other false online engagement |
| Disclosure | Failing to disclose to end users any known dangers of the AI system |
The Llama 2 and Llama 3 policies contain the same list, and the Llama 3.1 policy repeats it in a different format.[7][8][9]
The published policies still list military and warfare uses as prohibited, but Meta has made exceptions by announcement. On November 4, 2024 it said it was "making Llama available to U.S. government agencies, including those that are working on defense and national security applications, and private sector partners supporting their work", naming partners such as Lockheed Martin, Anduril and Palantir.[40] On September 23, 2025 it extended national-security access to the Five Eyes partners, France, Germany, Italy, Japan, South Korea, NATO and European Union institutions.[41] Neither announcement changed the text of the policies.[40][41]
Changes in the Llama 3.2 policy
The Llama 3.2 policy made the first substantive changes, and the Llama 3.3 and Llama 4 policies keep them.[10][11][12] It added a ban on intentionally circumventing or removing "usage restrictions or other safety measures" or enabling "functionality disabled by Meta"; extended the weapons clause to activities covered by the US Biological Weapons Anti-Terrorism Act of 1989 and the Chemical Weapons Convention Implementation Act of 1997; prohibited interacting with third-party tools or models designed to generate unlawful content, or presenting their outputs as associated with Meta or Llama; and reworded the personal-data clause.[10]
European Union exclusion for multimodal models
The Llama 3.2 policy also added this paragraph: "With respect to any multimodal models included in Llama 3.2, the rights granted under Section 1(a) of the Llama 3.2 Community License Agreement are not being granted to you if you are an individual domiciled in, or a company with a principal place of business in, the European Union. This restriction does not apply to end users of a product or service that incorporates any such multimodal models."[10] It covered the multimodal models released under the Llama 3.2 license, the 11B and 90B Vision models and Llama Guard 3 11B Vision, not the 1B and 3B text-only models.[10][39] The Llama 3.3 and Llama 4 policies repeat the paragraph with their own version names.[11][12] Meta described Llama 4 Scout and Maverick as natively multimodal, so the exclusion reaches the main Llama 4 models, as well as Llama Guard 4, a multimodal safety classifier distributed under the Llama 4 license.[25][12][42]
The restriction followed a statement Meta gave Axios in July 2024: "We will release a multimodal Llama model over the coming months, but not in the EU due to the unpredictable nature of the European regulatory environment."[19] Reporting at the time tied the decision to Meta's difficulty training on data from European users under the GDPR, after regulators had objected to its plan to use public Facebook and Instagram posts.[19] The clause is narrower than a regional ban: it removes the license grant from EU-based individuals and companies that would take the models as licensees, and it does not apply to end users of products built on them.[10]
Version comparison
| Agreement | Release or effective date | Commercial use | Outputs used to improve other models | Attribution | Derivative model naming | EU exclusion for multimodal models |
|---|---|---|---|---|---|---|
| LLaMA License | 2023-02-24 | No, non-commercial research only[14] | Outputs ("any data produced by the Software") limited to non-commercial research, like the model[14] | License copy and "LLaMA is licensed under the LLaMA License" notice | None | None |
| Llama 2 Community License | 2023-07-18 | Yes, unless over 700M MAU[1] | Prohibited, except to improve Llama 2 derivatives[1] | License copy and Notice file | None | None |
| Meta Llama 3 Community License | 2024-04-18 | Yes, unless over 700M MAU[2] | Prohibited, except to improve Llama 3 derivatives[2] | "Built with Meta Llama 3" and Notice file | Must begin with "Llama 3" | None[8] |
| Llama 3.1 Community License | 2024-07-23 | Yes, unless over 700M MAU[3] | Allowed, subject to naming rule[3] | "Built with Llama" and Notice file | Must begin with "Llama" | None[9] |
| Llama 3.2 Community License | 2024-09-25 | Yes, unless over 700M MAU[4] | Allowed, subject to naming rule | "Built with Llama" and Notice file | Must begin with "Llama" | Yes, in the policy[10] |
| Llama 3.3 Community License | 2024-12-06 | Yes, unless over 700M MAU[5] | Allowed, subject to naming rule | "Built with Llama" and Notice file | Must begin with "Llama" | Clause in the policy, but Llama 3.3 is text-only, so no model is affected[11] |
| Llama 4 Community License | 2025-04-05 | Yes, unless over 700M MAU[6] | Allowed, subject to naming rule | "Built with Llama" and Notice file | Must begin with "Llama" | Yes, in the policy[12] |
Derivative models
The agreements require anyone who distributes the Llama Materials or "any derivative works thereof" to provide a copy of the agreement, and the Acceptable Use Policy binds each user.[1][3] Fine-tuned versions of Llama models are therefore usually distributed with the matching community license, and some developers add terms of their own. Examples, as stated on their Hugging Face model cards:
| Model | Developer | Base model | License as stated by the developer |
|---|---|---|---|
| Code Llama | Meta | Llama 2 | Llama 2 license (Hugging Face tag)[31] |
| Vicuna v1.5 | LMSYS | Llama 2 | Llama 2 Community License Agreement[32] |
| Hermes 3 (8B to 405B) | Nous Research | Llama 3.1 | Llama 3 license (Hugging Face tag)[29] |
| Hermes 4 70B and 405B | Nous Research | Llama 3.1 | Llama 3 license tag; the 14B model, built on Qwen3, is Apache 2.0[29] |
| Tülu 3 | Ai2 | Llama 3.1 | Llama 3.1 Community License Agreement[28] |
| Llama-3.1-Nemotron-70B-Instruct | NVIDIA | Llama 3.1 70B | NVIDIA Open Model License, with the Llama 3.1 license listed as additional information and "Built with Llama"[30] |
| Llama-3.3-Nemotron-Super-49B-v1 | NVIDIA | Llama 3.3 70B | NVIDIA Open Model License, with the Llama 3.3 license listed as additional information[30] |
| DeepSeek-R1-Distill-Llama 8B and 70B | DeepSeek | Llama 3.1 8B and Llama 3.3 70B | Repository tagged MIT; the README says the models are derived from Llama and "originally licensed under" the Llama 3.1 and 3.3 licenses[35] |
The Hermes 4 family shows how a license follows the base model: the 70B and 405B versions, fine-tuned from Llama 3.1, carry the Llama license, while the 14B version from the same developer carries Apache 2.0.[29]
Dispute over the term "open source"
Meta's description
Meta introduced Llama 2 as "the next generation of our open source large language model".[17] In a letter published with Llama 3.1 in July 2024, Mark Zuckerberg compared Llama with Linux and wrote that "Meta is committed to open source AI."[23] The Llama 4 announcement used both terms, calling Scout and Maverick "open-weight natively multimodal models" and saying the release was made "in keeping with our commitment to open source".[25]
Open Source Initiative, 2023
Two days after the Llama 2 release, on July 20, 2023, OSI executive director Stefano Maffulli wrote that Meta was confusing open source with "resources available to some users under some conditions" and that OSI had asked Meta to correct its statement.[20] The post said the license failed points 5 and 6 of the Open Source Definition, which forbid discrimination against persons, groups or fields of endeavor, because paragraph 2 restricted commercial use for some users and the Acceptable Use Policy restricted use for some purposes. It also objected to the policy's bans on areas such as regulated substances and critical infrastructure, and to the general requirement to follow the law.[20] The dispute is one of the main examples discussed under openwashing.
Open Source AI Definition and Meta's reply, 2024
When OSI published the Open Source AI Definition 1.0 on October 28, 2024, Maffulli said that Google and Microsoft had agreed to stop using the term for models that were not fully open, but Meta had not.[24] A Meta spokesperson told TechCrunch: "We agree with our partner the OSI on many things, but we, like others across the industry, disagree with their new definition," adding that "There is no single open source AI definition, and defining it is a challenge because previous open source definitions do not encompass the complexities of today's rapidly advancing AI models."[24] The spokesperson defended the license and the Acceptable Use Policy as guardrails against harmful deployments and pointed to other efforts to define the term, including from the Linux Foundation and the FSF.[24]
FSF and OSI, 2025
On January 24, 2025 the FSF published its evaluation of the Llama 3.1 Community License, concluding: "This is not a free software license and you should not use it, nor any software released under it."[21] It also said the license "purports to hand over powers to the licensors that should only be exercised through lawmaking by democratically-elected governments."[21]
On February 18, 2025 OSI published "Meta's LLaMa license is still not Open Source", stating that "Llama 3.x is still not Open Source by any stretch of the imagination." It endorsed the FSF evaluation, listed failures of freedom 0 and Open Source Definition points 5 and 6, and said newer versions imposed more restrictions, "like excluding any persons in the European Union from using the model without explanation".[22] The policy text itself limits that exclusion to multimodal models and exempts end users of products that incorporate them.[10][12] The post said there was no need to apply the Open Source AI Definition to reach this conclusion, because the licenses already fail to grant the basic freedoms of the Open Source Definition for software.[22]
Treatment under the EU AI Act
The EU AI Act exempts providers of general-purpose AI models released under a "free and open-source licence" from some of its obligations, including parts of the documentation duties, unless the model has systemic risk.[27] In an April 10, 2025 letter to Commission Executive Vice-President Henna Virkkunen, a cross-party group of members of the European Parliament led by Birgit Sippel and Markéta Gregorová named Llama, said Meta forces anyone who builds a highly successful system on Llama "to negotiate a special licence with them", and urged the Commission to clarify that such systems "cannot be considered Open Source for the purposes of the AI Act".[26]
The Commission's guidelines on the obligations of general-purpose AI model providers, approved in draft on July 18, 2025 and adopted as a Commission communication dated November 19, 2025, set out its reading of the exemption. The license must allow access, use, modification and distribution; use may be limited only by attribution and same-terms requirements, apart from proportionate, safety-oriented restrictions. The guidelines list examples of restrictions that disqualify a license, including non-commercial or research-only limits, requirements to obtain separate commercial licenses for specific uses, and "usage restrictions triggered by user scale thresholds (e.g. requiring additional licensing if monthly active users exceed a certain number)".[27] They also say that safety and security measures on access, such as user verification, must not "unfairly discriminate against persons, for example on the basis of their country of origin", and that permitted safety terms must not "unfairly target specific categories of users".[27] The guidelines do not name any model or license; the user-threshold example describes the type of clause found in section 2 of every Llama Community License.[27][1]
Meta's later licenses
In 2026 Meta released open-weight models from Meta Superintelligence Labs under the Muse name, and did not use the Llama Community License for them. Muse Glimmer, released on August 10, 2026, is distributed under the Apache License 2.0, with a separate Muse Glimmer Usage Policy published in the repository.[36][33]
Muse Spark Open, released on October 15, 2026, uses a new "Muse Model License", which Meta described as having "limited commercial terms, which allow many developers to use the model for free".[34][38] Its structure differs from the Llama agreements:[34]
| Term | Llama Community License (3.1 to 4) | Muse Model License (Muse Spark Open) |
|---|---|---|
| Condition on large companies | Separate license needed above 700 million monthly active users on the release date | Notice to Meta and a separate written agreement needed only to offer the model as a "Model as a Service" by a licensee group with more than USD 20 million gross revenue in any 12 months; internal use and other commercial use are exempt |
| Branding | "Built with Llama" | "Built with Muse Spark Open" in the user interface or documentation |
| Derivative naming | Name must begin with "Llama" | No naming requirement; "Derivatives" include models improved with outputs |
| Usage policy | Acceptable Use Policy | Usage Policy attached as Exhibit A, with prohibited uses largely matching the Llama 3.2 to 4 policy, and a statement that the model is not intended for people under 18 |
| EU exclusion | Withholds multimodal models from EU-based licensees | None in the license or usage policy |
| Ownership of outputs | Not addressed (outputs appear only in the warranty disclaimer, the litigation clause and the naming rule) | "Meta claims no ownership of outputs generated by the Software." |
The Muse Spark Open repository on Hugging Face is gated, while the Muse Glimmer repository is public.[34][33] AI Wiki's open-weight license comparison and list of open-weight models compare these terms with other custom and standard licenses.
Continue reading
- LLaMALLaMA is the original name of a family of large language models developed by Meta AI.Mentioned 173 times above
- Open Source AI DefinitionThe Open Source AI Definition (OSAID) is a document published by the Open Source Initiative (OSI) that states what an artificial intelligence system must make available, and under what termsMentioned 4 times above
- Open weightsOpen weights is the practice of publishing the trained parameter tensors of a neural network so that anyone can download, run, fine-tune, and redistribute the model without obtaining a separate inference…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
- 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
- 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
More from See also
More to read
Topics:AI Policy & RegulationLarge Language ModelsOpen Source AI
References
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21Meta Platforms. "Llama 2 Community License Agreement" (Llama 2 Version Release Date: July 18, 2023). GitHub, meta-llama/llama-models. github.com/...LICENSE
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Meta Platforms. "Meta Llama 3 Community License Agreement" (Meta Llama 3 Version Release Date: April 18, 2024). GitHub, meta-llama/llama-models. github.com/...LICENSE
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Meta Platforms. "Llama 3.1 Community License Agreement" (Llama 3.1 Version Release Date: July 23, 2024). GitHub, meta-llama/llama-models. github.com/...LICENSE
- ^1 ^2 ^3Meta Platforms. "Llama 3.2 Community License Agreement" (Llama 3.2 Version Release Date: September 25, 2024). GitHub, meta-llama/llama-models. github.com/...LICENSE
- ^1 ^2 ^3Meta Platforms. "Llama 3.3 Community License Agreement" (Llama 3.3 Version Release Date: December 6, 2024). GitHub, meta-llama/llama-models. github.com/...LICENSE
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18Meta Platforms. "Llama 4 Community License Agreement" (Llama 4 Version Effective Date: April 5, 2025). GitHub, meta-llama/llama-models; also at llama.com/...license. github.com/...LICENSE
- ^1 ^2 ^3Meta Platforms. "Llama 2 Acceptable Use Policy". GitHub, meta-llama/llama-models. github.com/...USE_POLICY.md
- ^1 ^2Meta Platforms. "Meta Llama 3 Acceptable Use Policy". GitHub, meta-llama/llama-models. github.com/...USE_POLICY.md
- ^1 ^2Meta Platforms. "Llama 3.1 Acceptable Use Policy". GitHub, meta-llama/llama-models. github.com/...USE_POLICY.md
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Meta Platforms. "Llama 3.2 Acceptable Use Policy". GitHub, meta-llama/llama-models. github.com/...USE_POLICY.md
- ^1 ^2 ^3 ^4Meta Platforms. "Llama 3.3 Acceptable Use Policy". GitHub, meta-llama/llama-models. github.com/...USE_POLICY.md
- ^1 ^2 ^3 ^4 ^5 ^6 ^7Meta Platforms. "Llama 4 Acceptable Use Policy". GitHub, meta-llama/llama-models; also at llama.com/...use-policy. github.com/...USE_POLICY.md
- ^1 ^2Meta AI. "Introducing LLaMA: A foundational, 65-billion-parameter large language model". Meta AI blog, February 24, 2023. ai.meta.com/...large-language-model-llama-meta-ai
- ^1 ^2 ^3 ^4 ^5Meta Platforms. "LLaMA License Agreement" (2023), as reproduced in Wind River's license list. open.windriver.com/...llama-license-2023
- ^Meta. "llama" repository, llama_v1 branch: README (download request form) and LICENSE (GNU General Public License v3). GitHub. github.com/...llama_v1
- ^Vincent, James. "Meta's powerful AI language model has leaked online: what happens now?". The Verge, March 8, 2023. theverge.com/...age-model-llama-leak-online-misuse
- ^1 ^2Meta. "Meta and Microsoft Introduce the Next Generation of Llama". Meta Newsroom, July 18, 2023. about.fb.com/...llama-2
- ^Meta AI. "Introducing Llama 3.1: Our most capable models to date". Meta AI blog, July 23, 2024. ai.meta.com/...meta-llama-3-1
- ^1 ^2Dixit, Pranav. "Meta will withhold multimodal AI models from the EU amid regulatory uncertainty". Engadget, July 17, 2024. engadget.com/...d-regulatory-uncertainty-215543292
- ^1 ^2 ^3 ^4Maffulli, Stefano. "Meta's LLaMa license is not Open Source". Open Source Initiative, July 20, 2023. opensource.org/...ama-2-license-is-not-open-source
- ^1 ^2 ^3Siewicz, Krzysztof. "Llama 3.1 Community License is not a free software license". Free Software Foundation, January 24, 2025. fsf.org/...-license-is-not-a-free-software-license
- ^1 ^2 ^3Maris, Jordan. "Meta's LLaMa license is still not Open Source". Open Source Initiative, February 18, 2025. opensource.org/...license-is-still-not-open-source
- ^Zuckerberg, Mark. "Open Source AI is the Path Forward". Meta Newsroom, July 23, 2024. about.fb.com/...open-source-ai-is-the-path-forward
- ^1 ^2 ^3Wiggers, Kyle. "We finally have an 'official' definition for open source AI". TechCrunch, October 28, 2024. techcrunch.com/...al-definition-for-open-source-ai
- ^1 ^2 ^3Meta AI. "The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation". Meta AI blog, April 5, 2025. ai.meta.com/...llama-4-multimodal-intelligence
- ^1 ^2Sippel, Birgit; Gregorová, Markéta; and others. "Mislabelling of Open Source AI and consequences for enforcement of the AI act" (letter to Executive Vice-President Henna Virkkunen), April 10, 2025. Published by the Open Source Initiative. opensource.org/...commissionletterai.pdf
- ^1 ^2 ^3 ^4European Commission. "Guidelines on the scope of the obligations for providers of general-purpose AI models under the AI Act" (C(2025) 5045, approved July 18, 2025; Commission communication C(2025) 7719, November 19, 2025), section 4.2.1, paragraphs 73 to 81. digital-strategy.ec.europa.eu/...dels-under-ai-act
- ^1 ^2Allen Institute for AI. "Llama-3.1-Tulu-3-8B" model card, "License and use". Hugging Face. huggingface.co/...Llama-3.1-Tulu-3-8B
- ^1 ^2 ^3 ^4Nous Research. Model cards and license tags for "Hermes-3-Llama-3.1-8B", "Hermes-3-Llama-3.1-405B", "Hermes-4-70B", "Hermes-4-405B" and "Hermes-4-14B". Hugging Face, accessed November 2026. huggingface.co/...Hermes-3-Llama-3.1-8B
- ^1 ^2 ^3NVIDIA. "Llama-3.1-Nemotron-70B-Instruct-HF" and "Llama-3_3-Nemotron-Super-49B-v1" model cards, "License". Hugging Face. huggingface.co/...Llama-3.1-Nemotron-70B-Instruct-HF
- ^Meta. "CodeLlama-7b-hf" model repository (license tag llama2). Hugging Face. huggingface.co/...CodeLlama-7b-hf
- ^LMSYS. "vicuna-13b-v1.5" model card. Hugging Face. huggingface.co/...vicuna-13b-v1.5
- ^1 ^2 ^3Meta Superintelligence Labs. "Muse-Glimmer-30B" model card (license: Apache 2.0; USAGE_POLICY.md). Hugging Face, accessed November 2026. huggingface.co/...Muse-Glimmer-30B
- ^1 ^2 ^3 ^4Meta Superintelligence Labs. "Muse-Spark-Open-816B-A42B" model card and "Muse Model License" (version Muse Spark Open, release date October 15, 2026, with Exhibit A usage policy). Hugging Face, accessed November 2026. huggingface.co/...Muse-Spark-Open-816B-A42B
- ^DeepSeek. "DeepSeek-R1-Distill-Llama-70B" model card, section 7 "License". Hugging Face. huggingface.co/...DeepSeek-R1-Distill-Llama-70B
- ^Meta AI Research. "Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device". August 10, 2026. research.meta.ai/...use-glimmer-open-agentic-model
- ^Taori, Rohan, and others. "Stanford Alpaca: An Instruction-following LLaMA Model" (README, "Usage and License Notices"). GitHub, tatsu-lab/stanford_alpaca. github.com/...stanford_alpaca
- ^Meta Superintelligence Labs. "Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date". Meta AI Research, October 15, 2026. research.meta.ai/...muse-spark-open
- ^Meta Platforms. "Llama 3.2" and "Llama 3.2-Vision" model cards (MODEL_CARD.md and MODEL_CARD_VISION.md). GitHub, meta-llama/llama-models. github.com/...MODEL_CARD_VISION.md
- ^1 ^2Meta. "Open Source AI Can Help America Lead in AI and Strengthen Global Security". Meta Newsroom, November 4, 2024. about.fb.com/...-source-ai-america-global-security
- ^1 ^2Kaplan, Joel. "Strengthening US National Security by Making Llama Available to Key Allies". Meta Newsroom, September 23, 2025. about.fb.com/...king-llama-available-to-key-allies
- ^Meta. "Llama-Guard-4-12B" model card (license: Llama 4 Community License). Hugging Face. huggingface.co/...Llama-Guard-4-12B
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 · 4,952 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 112 claims against 44 sources incl. all Llama LICENSE and USE_POLICY texts; 8 minor fixes applied plus military-use exceptions and Llama Guard 4.
Cite this page: AI Wiki. "Llama Community License." aiwiki.ai, updated 17 Nov 2026, fact-checked 17 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/llama_community_license
