Sorry, but both of these benchmarks are terrible. First link compares run time of the whole programs, including start-up time, time for parsing file and printin

Sorry, but both of these benchmarks are terrible. First link compares run time of the whole programs, including start-up time, time for parsing file and printing out resulting matrix. I'm pretty sure matrix multiplication was the fastest of all these operations. I haven't read details of second benchmark (in fact, I've got lost in all these XXXFactories and couldn't find where things are actually happening), but I see no Python sources to compare with. So it's quite useless in this discussion too. Also note, that performance was not the only point of my comment. In fact, my point was in simplicity of designing complex scientific systems. In Python you have excellent stack: NumPy + SciPy + Scikit-* + Pandas + Numba + Cython + Theano + Pylearn2 +... The list is endless, and all these tools may easily be used together. In R you have CRAN with everything you may ever be wanting. In Matblab you have a number of both - open-source and proprietary libraries. Julia community is also growing quickly. (Note, none of these libraries uses OOP as its primary paradigm.) And what about Java? Java has a little disjoint set of libraries that neither solve large set of problems, nor interoperate with each other well.