+1 on this. I really like the visualisation. I would love a 4th column which summarises the implication.

+1 on this. I really like the visualisation. I would love a 4th column which summarises the implication. Reproducibility: Open Source enables it — if code, training data, and configuration are all published — while open weight alone does not.

Like with a closed source library you can build on top of an open-weight model — even inspect or fine-tune the weights — but you cannot see how it was trained or rebuild it from source. That is the fundamental difference: access to the artifact versus access to the recipe.