What it means
An open-weight model is an AI model whose trained parameters, called weights, are published so that others can download, run and often adapt it. The licence may still restrict use, and the training data and training code are often not published.
Real-world example
A university hospital wants to test AI summarisation of discharge letters without sending patient data outside its network. It downloads an open-weight model, checks the licence permits use in its context, and runs it on its own servers with access limited to the project team.
Why it matters for AI adoption
Open-weight models enable on-premises use, customisation and less dependence on a single vendor. But "open" in the name does not mean the model is free to use for any purpose, that its training data is known, or that it is safe. Licence, documentation, security updates and evaluation all need checking.
Often confused with
Open-source models
"Open source" traditionally means users can use, study, modify and share the software freely. For AI, the Open Source Initiative's definition also expects enough information about training data and code to reproduce or substantially modify the model. Many models described as "open" are open-weight only: the weights are available, but training data and code are not, and licences may restrict use.
Questions to ask
- What does the licence allow, including for public-sector or commercial use?
- What do we know about how the model was trained and evaluated?