An open-weight artificial intelligence model distributes its learned weight files so users can host, inspect, quantize or adapt the model themselves. The release may also include code and documentation, but access and reuse depend on the license.
Open weights provide deployment control and local inference options without automatically making the training data or full development process open. Hardware requirements, safety controls, support and total operating cost still need evaluation.
