Open model licences explained: what Apache 2.0 and MIT mean for you
Apache 2.0 and MIT are permissive open licences. If you simply use a model under either one on your own laptop, including for paid work, you do not have to pay anyone or ask permission. The conditions mainly apply if you pass the model on to others: keep the copyright and licence notices with it (and, for Apache 2.0, say what you changed). Neither licence promises the model is accurate or fit for a purpose. Some other popular models, such as Gemma and Llama, use their own licences with extra rules.
Open AI models are free to download, but "free to download" is not the same as "free to do anything with". Each model comes with a licence, a short legal document that says what you may do. The two you will meet most among small, laptop-sized models are Apache 2.0 and MIT. Here is what they mean for an ordinary user, without the legalese.
Why a model has a licence
A model is a file that someone spent a lot of money to make. The licence says how they are sharing it. Some publishers share models under well-known open source licences. Others write their own terms with extra rules. You can see a model's licence on its Hugging Face page, in the model card and usually in a LICENSE file.
Apache 2.0 in plain English
The Apache License 2.0, from the Apache Software Foundation, gives you:
- Free use for any purpose, personal or commercial, worldwide, with no fee.
- The right to copy, change and share the model, including as part of something you sell.
- A patent licence from the people who contributed to it, covering their contributions to it. That patent licence ends if you start a patent lawsuit claiming the model itself infringes a patent.
If you pass the model on to someone else, you must:
- Include a copy of the licence.
- Keep the copyright, patent and attribution notices, including any NOTICE file.
- Mark any files you changed.
It does not give you rights to the publisher's name or trademarks, and it comes with no warranty: the model is provided "as is", and the publisher is not liable if it gets something wrong.
MIT in plain English
The MIT License is one of the shortest licences around. It lets anyone use, copy, change, merge, publish, share, sublicense and sell the work, free of charge. The one condition: keep the copyright notice and the permission notice with all copies. Like Apache 2.0, it comes "as is", with no warranty. Microsoft's Phi-4-mini-instruct is an example of a small model published under MIT.
Side by side
| Apache 2.0 | MIT | |
|---|---|---|
| Use at work, for money | Yes | Yes |
| Change it | Yes, mark changed files | Yes |
| Share it | Yes, with licence and notices | Yes, with notices |
| Explicit patent licence | Yes | Not mentioned |
| Use of the publisher's trademarks | Not granted | Not mentioned |
| Warranty | None | None |
| Usage rules (what you may use it for) | None | None |
Other model licences
Some well-known small models use their own licences instead. Google's Gemma models are published under the Gemma terms of use, and Meta's Llama models under a Llama Community License. Both let you use the models for many purposes, but add rules of their own, such as a list of uses that are not allowed, and in Llama's case extra conditions for very large companies. They can be perfectly fine to use, but they are not the same as Apache 2.0 or MIT, so read them before relying on them at work.
How WorkOffline handles this
WorkOffline only offers models under Apache 2.0 or MIT. The three recommended models that come with every ticket are all Apache 2.0: Lite is Qwen3 4B Instruct 2507, Standard is Qwen3 8B, Large is Qwen3 14B, and the search helper is nomic-embed-text v1.5, each as stated on its publisher's model card. The Pro model library also includes MIT models, such as Microsoft's Phi-4 mini and Phi-4. The app shows each model's licence when you choose it, and downloads the files from our own Hugging Face organisation, pinned to an exact version and checked against a fingerprint before use.
Every ticket includes chat with your files and writing help, from one flat white. See tickets and prices.
Frequently asked questions
Can I use an Apache 2.0 AI model for commercial work?
Yes. Apache 2.0 allows use for any purpose, including commercial work, without paying a fee. If you pass the model on to others, keep the licence and notices with it.
Do I own what an open model writes for me?
Apache 2.0 and MIT cover the model itself and put no special conditions on the text it writes for you. Who owns AI-written text is a separate legal question that varies by country, so take advice if it matters to you.
What is the difference between Apache 2.0 and MIT?
Both are permissive. MIT is very short: keep the copyright and permission notice. Apache 2.0 is longer and adds an explicit patent licence, asks you to mark files you changed, and makes clear it gives no rights to the publisher's trademarks.
Is "open weights" the same as open source?
Not always. Open weights means you can download the model file. Whether you can use it freely depends on the licence. Apache 2.0 and MIT are recognised open source licences; some model licences add usage rules that open source licences do not have.
Is this legal advice?
No. This is a plain-English summary to help you understand the basics. Read the licence itself, and ask a lawyer if you plan to redistribute a model or build a product on one.
Sources
- Apache License, Version 2.0 (Apache Software Foundation)
- The MIT License (Open Source Initiative)
- Qwen3-4B-Instruct-2507 model card (Hugging Face): 4.0B parameters, Apache 2.0
- Qwen3-8B model card (Hugging Face): 8.2B parameters, Apache 2.0
- Qwen3-14B model card (Hugging Face): 14.8B parameters, Apache 2.0
- nomic-embed-text-v1.5 GGUF (Nomic, Hugging Face): Q8_0 file 146,146,432 bytes, Apache 2.0
- Phi-4-mini-instruct model card (Microsoft, Hugging Face): MIT licence
- Phi-4 model card (Microsoft, Hugging Face): MIT licence
- Gemma 3 4B model card (Google, Hugging Face): Gemma licence
- Llama 3.2 3B Instruct model card (Meta, Hugging Face): Llama 3.2 Community License
An evergreen guide from WorkOffline. No tracking scripts on this page.