You may use, change, sell and redistribute an Apache 2.0 model, including inside a commercial product, as long as you pass on the licence text and the required notices. Unlike MIT, Apache 2.0 also gives you an explicit patent licence from each contributor.

This summarises the licence text. It is not legal advice.

What you may do

Section 2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce the work, prepare derivative works, publicly display and perform it, sublicense it and distribute it, in source or object form. For a model, that covers running the weights, fine-tuning them, quantising them and shipping the result inside your own product.

Section 3 adds a patent licence. Each contributor grants you a licence to make, have made, use, offer to sell, sell and import the work, for patent claims that their contribution necessarily infringes alone or by combination with the work to which it was submitted.

Section 4 lets you add your own copyright statement to your modifications and offer them under different licence terms, provided your use and distribution of the original work still follows Apache 2.0. A closed-source product built on an Apache 2.0 model is allowed.

What you must do

If you redistribute the work or a derivative, Section 4 sets four conditions:

  1. Give every recipient a copy of the licence.
  2. Mark any files you changed with prominent notices that you changed them.
  3. Keep all copyright, patent, trademark and attribution notices in the source form of the derivative works you distribute.
  4. If the work ships a NOTICE file, any derivative works you distribute must include a readable copy of the attribution notices in it, excluding notices that do not pertain to any part of the derivative works. Put them in a NOTICE file you distribute, in the source or documentation, or in a display your software generates where third-party notices normally appear. The NOTICE contents are informational and do not modify the licence.

There is no duty to credit the licensor on a website or in a user interface. The duty is to preserve notices when you redistribute.

What you may not do

Section 6 grants no right to use the licensor’s trade names, trademarks, service marks or product names, except for reasonable and customary use in describing the origin of the work. Calling your fine-tune by the original brand name is not covered.

Under Section 3, if you start patent litigation alleging that the work infringes a patent, the patent licences you received for that work end on the date you file.

Sections 7 and 8 disclaim warranties and limit liability. You take the model as it is.

How it compares with MIT

MIT is shorter and has one condition: keep the copyright and permission notice. Apache 2.0 adds the express patent licence, a duty to flag modified files and a trademark clause. For most teams the practical gap is small. The Apache 2.0 vs MIT comparison puts them side by side.

Check the repository, not only the label

A licence badge on a model page covers what that repository says it covers, and derived models can carry the base model’s terms. The DeepSeek-R1 model card says its Qwen-based distilled models are derived from the Qwen 2.5 series, originally under Apache 2.0, and its Llama-based ones carry Llama licences. Read the model card and the LICENSE file for the exact repository you download.

Models on this licence

Example open-weight models from our catalogue that carry Apache 2.0 are listed after the FAQ.

FAQ

Can I use an Apache 2.0 model commercially?

Yes. Sections 2 and 3 grant rights to use, modify, distribute and sell, with no revenue or user-count limit in the licence text.

Do I have to release my fine-tuned model under Apache 2.0?

No. Section 4 lets you add your own licence terms to your modifications, provided your use and distribution of the original work still follows Apache 2.0.

Do I have to show a credit in my app?

No on-screen credit is required. You must keep the notices and include the licence text when you redistribute the work or a derivative.

Does Apache 2.0 cover the training data?

It covers the work the licensor distributes, which is the model and any code in the repository. It says nothing about the training data; the model card is the place to check.