I suspect part of the reluctance to release Open Source models is the fact that they’d have to disclose what data was used to train it