Open-Weight Model

An open-weight model is released with the numerical parameters learned during training, allowing others to run it on their own infrastructure and often to fine-tune or study it. Open weights can improve portability, research access, customization, privacy control, and independence from a hosted API. The term does not automatically mean open source: the training code, dataset, architecture details, or commercial rights may still be restricted by a license. Users must review those conditions as well as security and safety implications. Local control also transfers responsibility for infrastructure, updates, monitoring, misuse prevention, and compliance from a managed provider to the organization operating the model.