I have a pet peeve in this industry, and it is this sentence: "the model is open source." Half the time, it is not.
"Open weights" and "open source" are not the same thing, and mixing them up is how licensing mistakes get made. Open weights means you can download the model parameters. That is the whole definition. It says nothing about what you are allowed to do with them.
Open source, properly speaking, means the license meets the Open Source Definition: free use, modification, redistribution, no discrimination against fields of use. A lot of "open" models fail this test because their licenses restrict certain uses or impose user thresholds. They are open weights with strings attached. AI model license explained in one line: the label is marketing, the license file is the contract.
Why does this distinction matter in practice? Procurement. Enterprise questionnaires now ask about it, and if your answer to "is this open source?" is "well, the weights are downloadable," you will get follow-up questions. Answer precisely the first time: name the license, link the file, move on.
Call things what they are: open weights with a specific license. Then evaluate that license on its own terms. An open source AI license database search tool is only useful if it makes this distinction instead of lumping everything under "open." And an open source model license comparison database that cannot tell you which "open" models fail the OSI test is just a catalog.
My practical advice: stop shopping by label, start shopping by license text. Two models both marketed as "open" can have wildly different terms. And keep a dated copy of any custom license. Save the text on the day you download the weights. If terms change later, you can show what applied when you shipped. Version control is not just for code.
I compare the actual license text of 200 models at localaitools.fyi in plain language, and I write commercial-use clearance reports for teams shipping on open weights.