Do you mean for a given model structure, can two sets of weights give substantially the same outputs?
Even if that were possible, it would be suspicious if you were to release an open model whose model architecture is identical to that of a closed one from a competitor.
If that is what happened, we'd know about it by now.
