AI glossary · Beginner · AI platforms & solutions · Safety & alignment
Model license
A model license is the set of usage rules that defines who may use, modify and redistribute an AI model's weights, for what purposes and under what conditions. Because it determines in particular whether commercial use is allowed, it should be the first thing you check when choosing a model.
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Infrastructure & using models
- Added
- 2026-10-12
What types are there
It is easiest to think of three broad groups. Permissive open-source licenses such as Apache 2.0 and MIT broadly allow commercial use, modification and redistribution, with little more than a notice requirement. Some companies create their own community licenses that allow commercial use but attach conditions such as user-scale limits, name attribution and prohibited uses (usage policies). Licenses with non-commercial terms, such as CC BY-NC, allow use only for research and personal purposes.
What each license type allows
| Commercial use | Modification and redistribution | Additional conditions | |
|---|---|---|---|
| Permissive (Apache 2.0, MIT) | ✓ | ✓ | △ |
| Company community license | △ | △ | ✓ |
| Non-commercial (CC BY-NC, etc.) | ✕ | △ | ✓ |
Where you'll see it
Check the 'License' field in the info panel on the right of a Hugging Face model page, the LICENSE file in the repository, and the body of the model card. Some models are 'gated', meaning you must agree to the terms before downloading the files. For models used only through the cloud or an API, the terms of service play the role of the license. Training data sources and output usage conditions are sometimes stated separately, so read them as well.
Good to know
The most common misconception is assuming commercial use is unrestricted just because something is called 'open'. Even open-weight models may prohibit commercial use or attach conditions, depending on the license. Derived models that are fine-tuned, quantized or merged usually inherit the original license terms, so check whether the license stated on community-uploaded variants matches the original. Also, the license of software such as an inference engine is separate from the license of the model weights run on it. It is safest to go through a legal review before adopting a model in-house.
Example
Even when an inference engine is released as open source, like Google AI Edge's 'ML Drift' featured on this site, the license of the model weights you run with that engine must be checked separately. In practice, a common scene is comparing the License field on Hugging Face for a few candidate models and dropping those with non-commercial terms from the internal service shortlist first.