Lets say I have 2 different versions of QAOA for the same graph. Maybe there's a difference in number of layers, or the classical optimizer itself between these two versions. Version 1 gives me a maxcut of 5 and Version 2 gives me a maxcut 7. Can I guarantee that Version 2 is performing better because of the larger maxcut value? Can I claim that Version 2 is better since it gives me higher value?

You can say that for this specific graph, version 2 performed better than version 1. However, in order to generalize that statement to "version 2 is better than version 1", you would need to compare them over a broad set of instances.