Yes this is precisely it! And you’re right that Meta gets credit for releasing its model. It would be even better if it became possible for an open source project to train its own new models explicitly for open release …. Pricey but maybe possible?