Python. Hands down.

Oh! I don’t mean to say that MATLAB is useless for ANN work: it is not useless at all! But MATLAB was created to be a general mathematical engine for engineering and statistics purposes, and it does that job very well indeed.

Python, on the other hand, was created as a general-purpose computer language, easy to use and to program with. And it does that very well; including being able to interface and take full and efficient advantage of ANN subsystems (libraries, if you want to call them that) designed for functionality, correctness and—particularly—efficiency and speed. Those libraries (the best known are TensorFlow and PyTorch) have quickly become a de-facto standard and are used almost everywhere, including many thousands of citations in research papers that use them to achieve the reported results. Furthermore, those libraries are 100% open source, meaning any researcher can use them with no “strings attached,” and can also extend and specialize them so that others can take advantage of the new research results. (Collected)

If I needed to teach the basics of ANN design, I would consider using Matlab because it has a graphical user interface. The GUI for both machine learning (in Statistics and Machine Learning Toolbox) and deep learning (in Deep Learning Toolbox) is rather simple and can be used easily to explain basics even to students who are not really fond of programming.

If you are asking about some serious work, please refer to the excellent answer by Mostafa Sameer

I´ve enjoyed Simulink like no other, but don´t think it offers a toolbox for DL.

Python is much better than Matlab to design deep learning models