noun · AI concept
Open-weight model
also called open weights
A model you can download and run yourself. You get the finished brain, though not necessarily the recipe it was made with.
A model whose trained weights are publicly released so they can be downloaded and run outside the original developer's service.
The trained weights are published so anyone can run, test and often fine-tune the model on their own hardware. That is different from open source, since the training data, training code and usage license may be withheld or restricted.
Real-life example
Your company cannot send customer data to an outside service. You download an open-weight model and run it on your own servers, so the data never leaves. You now also own the hardware, updates and monitoring that a hosted service would have handled.
Where the analogy breaks
-
Open weights are not the same as open source. You may get the model but not the data or training code, and the license can limit what you do.
-
Running it yourself has costs. You pay for hardware and upkeep, and you handle scaling, updates and safety checks.
-
Quality varies a lot. Some open models match hosted ones on certain tasks and others fall well behind, so test on your own work.