Model weights encode the numerical settings learned during training. An open-weight release provides those settings rather than offering access only through a hosted service. Depending on the license and available software, a user can run the model on their own infrastructure or adapt it for a specific task.
Access to weights does not necessarily include training data, unrestricted usage rights or a complete reproducible training process. It also does not remove hardware requirements. A very large model may require substantial memory and computing resources even when its weights are freely downloadable.
ELI5
A model provider can make the trained model available for other people to run rather than offering access only through its own service. Those downloadable learned values are called weights.
For example, an organization might run an open-weight writing assistant on its own computers. It still needs to check the license, provide enough computing power and decide what information the assistant may access.
Is open-weight the same as open source?
No. Available weights do not by themselves establish open-source licensing or access to the complete code, training data and training process.
Can every open-weight model run on a laptop?
No. Feasibility depends on the model's size, memory requirements, software and the laptop's hardware, not merely whether the weights are available.
