From my perspective, R is great for certain tasks (more related to classical machine learning and statistical analysis as you have mentioned). There are several libraries that have demonstrably better implementations in R than Python (lme4 comes to mind).

I will have to circle back to the same point regarding deep learning and Python. Whilst R does offer deep learning capabilities in its modern iterations, the uptake of using R in such a fashion pales in comparison to the uptake of Python, so the same comment I made about MATLAB still applies.