Both propriety and open weight models are closed source. Open source would include training data so anyone could train a similarly capable model.