Nicer syntax for mathematical operations, but it's very slow (even slower than Python, similar to R) and it doesn't always work with Matlab code like it's supposed to.
Julia is just a straightforward improvement (substantially faster, but it has the nicest mathematical syntax of anything I've seen).
Compared to numpy, octave has a nicer syntax for matrix math, but a bit idiosyncratic as a general language. It doesn't have anywhere near the ecosystem of Python/NumPy/SciPy. It is also nice in that there is a standard development shell with REPL plotting, debugging and text editor. Like numpy it is pretty slow if you have to actually use for loops over large data sets.