open-weight model

An open-weight model is one whose trained parameters, its model weights, are made available to download, so anyone can run the model on their own computers or servers. This is in contrast to closed models, which you can only use through a provider’s cloud.

The appeal is control. With an open-weight model you can keep your data in-house, avoid per-message fees, run offline, and customise the model through fine-tuning. That makes open-weight one of the most useful terms to know when reading AI news.

“Open-weight” is not quite the same as “open source”. You get the finished model to run, but not necessarily the training data or full recipe, and the licence may place limits on use. Even so, capable open-weight models from labs like Mistral, MiniMax and others are steadily closing the gap with the big closed providers.