An open-weight artificial intelligence model distributes the numerical weights produced during training. Developers can usually run the model on their own or third-party infrastructure and may be able to fine-tune it, depending on the license and available tooling.
Open weights do not necessarily mean that training data, training code, evaluation details, or every usage right is open. They do, however, make the model portable across compatible inference providers and reduce dependence on a single hosted application programming interface.
