Open Weights vs Open Source
The distinction that keeps lawyers, researchers, and practical developers arguing in separate rooms.
Open weights means the trained model parameters are available. You can often download and run the model, subject to the license.
Open source, in the stricter software sense, means the source code is available under an open-source license. For AI models, people often also care whether the training code, training data, recipes, and evaluation details are open.
Those are not the same claim.
Why it matters
If a model is open weight but has a restrictive license, your company may still have limits on commercial use, redistribution, or scale.
If the training data is not public, researchers cannot fully reproduce the model.
If the training code is not available, you may not know how the final behavior was produced.
The practical questions
Before using a model at work, ask:
- Can we use it commercially?
- Can we modify it?
- Can we host it for customers?
- Do we need to attribute it?
- Are there usage restrictions?
"Open" is not enough. Read the license.