Is this kind of thing also possible with one of these self-hosted models in a comparable way, or are they mostly good for coding?

To me, and I think to many other people as well, the language most suited for anything data-related is Python, not R. I might be wrong, but If I am I won't know it after reading this article, because it doesn't compare R to other options on the table. R is only the best at something if it offers advantages over the other options, and to be honest I very much doubt that this is the case when comparing against Python.

Anecdotally, a number of years ago at university I took a class titled "Statistical programming with R" because I had heard good things about it and was looking forward to a chance to learn a new tool. Unfortunately I learned pretty quickly that I had to fight R every step of the way to get it to do what I wanted. Everything seemed arcane, convoluted, and complicated. Went back to Python and never looked back. I don't doubt that one can do great things with R, but the effort needed to get there simply doesn't seem worth it to me when Python seems so much more accessible.

Having said all that, I would be quite interested in a comparison of typical data science (or data journalism) tasks in both R and Python by someone who is good at both. After having read the article headline I had hoped it went into that direction. I was disappointed to see that it's essentially just a statement of opinion that isn't backed up in any meaningful way.