Frequently Asked Question

What is the difference between open source AI and open weights?

An open weights release publishes a model's trained parameters for download. You can run the model, inspect it, and fine-tune it. That is a real and significant freedom, and it is what makes local inference possible.

It is not the same as open source. A weights release typically omits the training data and the training code, and sometimes carries a licence restricting commercial use or field of application. You can run the artefact but you cannot rebuild it from source, which by the standard the free software movement applied to compilers and kernels makes it closer to a freely redistributable binary.

The Open Source Initiative published an Open Source AI Definition in October 2024 requiring enough information about training data for a skilled person to reconstruct a substantially equivalent system. Several widely used open-weight models do not meet it. The terminology remains contested, so when you read that a model is "open", check what was actually released.