Mozilla is highlighting a new framework for understanding open source AI, arguing that model weights, training data, code, documentation, and other components can have very different levels of openness.