Open-weight model
An AI model whose trained parameters are released publicly, so anyone can download, run, and adapt it on their own hardware. Open-weight models offer control and privacy that closed, API-only models cannot, and they anchor a large community of builders.
When a model is trained, its knowledge ends up stored in millions or billions of numbers called parameters, or weights. Many leading models keep these weights private and let you use the model only through an online service. An open-weight model does the opposite: the weights are published, so anyone can download the model, run it on their own machines, inspect it, and fine-tune it for their own purposes.
This matters for reasons of cost, control, and privacy. An organisation handling sensitive data — a hospital, a bank, a law firm — may prefer to run a model entirely on its own infrastructure rather than send information to a third party. Open-weight models also let researchers and startups build on strong foundations without paying to train from scratch, which has made them the backbone of a large and fast-moving developer community.
A note on wording: “open-weight” is not always the same as “open source”. Releasing the weights lets people use and adapt a model, but the training data and full recipe are often still withheld, so a model can be freely downloadable without being open in the fuller sense the software world means by the term.