Internet Is Fun 
“Open source” meant something specific for thirty years: you can read the source, change it, and share your changes. Applied to AI models the phrase has become much looser, and the difference matters if you are planning to depend on one.
Three things the label is used for
Open weights. The trained model can be downloaded and run on your own hardware. This is the most common meaning, and it is genuinely useful: nobody can switch it off, change it under you, or read what you send it.
Open weights with conditions. Same, but the licence restricts something — commercial use above a certain size, competing products, particular applications. Still valuable, but it is a licence you have to actually read.
Genuinely open source. Weights, training code, and enough detail about the data to reproduce the work. This is rare, because the data is usually the part nobody wants to publish.
Why the distinction is practical, not pedantic
If a model is open-weights you can run it offline, which settles most privacy questions on its own. If it is open-weights-with-conditions you need to check whether your use is allowed before you build on it. If it is fully open you can audit what went in, which is the only version where claims about bias or training data can be checked rather than taken on trust.
All three are described as “open” in announcements. Only one of them lets you verify anything.
What to check before relying on one
- Can you download the weights, or only call an API?
- What does the licence actually restrict — and does your use fall inside it?
- Is there any published detail about the training data?
- Can it run on hardware you have, or does “open” still mean renting a datacentre?
The reasonable position
Open weights are a real gain even without open data. Being able to run a model locally, indefinitely, on your own machine is the difference between using a tool and renting access to one.
Just do not read “open source” as a guarantee of transparency. It usually means “you can have the finished thing”, not “you can see how it was made”. If you want to try some, our AI tools list includes where the open models actually live, and this guide covers judging whether any of them are worth your time.
