Local AI

Open-Weight AI Model Licenses Explained: Apache 2.0, MIT, Llama and Gemma Terms

By Web Asset House Studio · October 28, 2026

Note: Some future links on this site may be affiliate links. We never claim hands-on testing; guides are researched from the public sources listed in each article.

How we wrote this: A summary based on the license texts themselves and Hugging Face's documentation (linked below). This is general information, not legal advice. Licenses change between model versions, so always read the license that ships with the exact model you download.

Running a model on your own computer feels like owning it, but every model you download comes with a license. Some are standard open-source licenses. Others are custom agreements with conditions on naming, distribution or use. If you use local AI in a business, a five-minute license check is worth doing.

Where to find the license

On Hugging Face, the license is declared in the model card's metadata. Hugging Face's license documentation lists the identifiers you'll see, from standard ones like apache-2.0 and mit to model-specific ones like llama3.1 and gemma. When a repo says license: other, the full text should be in a LICENSE file in the repo. Read that file.

Standard permissive licenses

Apache 2.0. The Apache License 2.0 grants a perpetual, worldwide, royalty-free copyright license and a patent license. If you redistribute the work, you must include a copy of the license, mark files you changed, keep existing notices, and pass along any NOTICE file. The patent license ends for anyone who sues claiming the work infringes a patent. It doesn't grant trademark rights. Examples:

OpenAI's gpt-oss models are released under Apache 2.0.
Google publishes the Gemma 4 license as Apache 2.0.

MIT. The MIT License is shorter still: you can use, copy, modify and sell, as long as you keep the copyright and permission notice. OpenAI's Whisper speech model is MIT-licensed (see our private transcription guide).

Custom model licenses

Llama (Meta). The Llama 3.1 Community License is free of royalties but has conditions. A few that matter most:

If you redistribute Llama materials, or a product that contains them, you must include the agreement and prominently display "Built with Llama".
If you use Llama, or its outputs, to create or improve a model you distribute, the model's name must start with "Llama".
Use must follow Meta's Acceptable Use Policy, which the license incorporates.
Companies whose products had more than 700 million monthly active users on the release date must request a separate license from Meta.

Other Llama versions have their own license files, so check the one that matches your model.

Earlier Gemma models (Google). Models listed on the Gemma Terms of Use page, including Gemma 1 to 3 and Gemma 3n, use custom terms. Redistributors must pass along the use restrictions, give recipients a copy of the terms, mark modified files, and include a NOTICE file. Use must comply with the Gemma Prohibited Use Policy. The terms say Google claims no rights in the outputs you generate.

A quick checklist before business use

Exact model, exact license. Two versions of the "same" family can have different terms. Gemma 3 and Gemma 4 are an example.
Are you redistributing? Using a model internally is different from shipping it inside an app or offering it as a hosted service. Most notice and naming duties kick in when you distribute.
Acceptable-use policies. Custom licenses often incorporate a use policy by reference. Read it.
Fine-tunes and merges. A community fine-tune generally still carries its base model's license conditions, so check the base model's license as well as the uploader's card.
Keep a record. Note the model name, version, source URL and license in your project files.

Private AI Computer's model pages list memory needs for popular open models, so you can shortlist models that fit your hardware before reading their licenses.