Meta’s Community Licence allows commercial use of Llama, with conditions. A company whose products had more than 700 million monthly active users in the month before a version’s release date needs a separate licence from Meta. Llama 3 and later require “Built with Llama” style credit and model naming. Every version binds you to Meta’s acceptable use policy.
This summarises the licence text. It is not legal advice.
What you may do
Each version grants a non-exclusive, worldwide, non-transferable and royalty-free limited licence to use, reproduce, distribute, copy, create derivative works of and modify the Llama materials. You can fine-tune, quantise, host and ship products built on the models. If you receive Llama materials from a licensee as part of an integrated end-user product, the commercial-terms section does not apply to you.
What you must do
When you distribute the materials or derivatives, every version requires you to give recipients a copy of the agreement and to keep a “Notice” text file with a stated attribution line. You must also follow applicable law and Meta’s Acceptable Use Policy, which the licence incorporates by reference.
The 700 million user clause
Every version has the same test. It counts the monthly active users of the products or services made available by or for you or your affiliates. If that number “is greater than 700 million monthly active users in the preceding calendar month”, measured on the version’s release date, you must request a licence from Meta, which Meta “may grant to you in its sole discretion”. Until Meta grants it, you are not authorised to exercise the rights. The test applies at one date; it is not an ongoing limit on usage.
Differences between versions
Llama 2 (release date 18 July 2023) asks only for the Notice file as attribution, with no “Built with” display rule and no naming rule. It does bar using the materials or their output to improve any other large language model, other than Llama 2 and its derivatives.
Llama 3 (18 April 2024) adds a display rule: show “Built with Meta Llama 3” on a related website, user interface, blog post, about page or product documentation. If you use the materials to create, train, fine-tune or improve an AI model that you distribute, its name must begin with “Llama 3”. The ban on improving other models remains.
From Llama 3.1 (23 July 2024), through 3.2, 3.3 and 4, the display text becomes “Built with Llama” and model names must begin with “Llama”. The naming rule now reaches models trained with the materials or their outputs. In the texts we checked, the clause barring use of outputs to improve other models no longer appears. Llama 3.2 is dated 25 September 2024, Llama 3.3 is dated 6 December 2024 and Llama 4 has an effective date of 5 April 2025.
The Llama 3.2 and Llama 4 multimodal models carry an EU carve-out. Their acceptable use policies state that the Section 1(a) rights are not granted to an individual domiciled in, or a company with its principal place of business in, the European Union for multimodal models. End users of a product that includes such a model are not affected.
What you may not do
You may not use Meta’s names or marks except as the licence allows, which includes the “Llama” mark for the naming rule on later versions. If you sue Meta alleging that the materials or their outputs infringe your intellectual property, your licences end on the date you file. Meta may also terminate the agreement if you breach it. The licence is governed by California law.
How it compares with the nearest licence
The Gemma Terms of Use are the closest comparison: a custom licence with a prohibited use policy and distribution conditions, but no user-count threshold and no naming rule. Against Apache 2.0, Llama adds the 700 million clause, the credit and naming rules, and the policy.
Models on this licence
Example Llama models from our catalogue, one for each licence version, are listed after the FAQ.
FAQ
Is Llama open source?
The weights are published, but the licence adds conditions: the 700 million user test, credit and naming rules, and an acceptable use policy. Meta calls it a Community License.
Can a startup use Llama commercially?
Yes. The 700 million test looks at the monthly active users of your products and your affiliates’ on the release date, and the other conditions are credit, naming and policy compliance.
Do I have to name my fine-tuned model “Llama”?
On Llama 3 and later, yes, if you distribute or make available an AI model created or improved with the materials. Its name must begin with “Llama 3” (Llama 3) or “Llama” (3.1 onward).
Can I use Llama output to train another model?
Llama 2 and Llama 3 bar it, except for Llama derivatives. The later texts we checked do not contain that clause, but they do apply the naming rule.