Open-weight model

Publishing weights lets organisations run a model on their own servers, in a chosen cloud region or on local devices, and adapt it with fine-tuning. Licences vary widely: some permit almost any use, while others restrict commercial use, user numbers or certain applications. The Open Source Initiative's Open Source AI Definition sets a stricter bar that also covers information about training data and code, which many open-weight releases do not meet.

Why it matters: open-weight models give buyers more control over data residency, privacy, cost at scale and vendor lock-in. Many third-party products may also run an open model behind the scenes. The trade-off is that you or your provider take on the hosting, security, updates and safety tuning that a closed-model vendor would otherwise handle.

Example: a hospital runs an open-weight language model on its own servers so patient records never leave its network.

Limits: running large models requires significant GPU capacity, and the capability gap between open-weight and leading closed models varies over time. Always read the licence before commercial use.