I think this is a good question because I don't really know what "letters" means in the world of mathematical publishing. While I have a paper published in Math Research Letters, I am having trouble pointing to a single way in which that paper is different from any of my other papers: it is slightly short (12 pages) but I have several other papers which are shorter, it contains complete proofs, the syle is not especially conversational or different from the norm...
With regard to Comptes Rendus: it still exists, and it is still a very high quality journal, so far as I know. I do not have a CR paper, and I wish I did, but setting aside the limitations of my own research achievements I am not sure exactly what papers to submit to this excellent and highly French journal. My understanding though is that it is roughly analogous to Proceedings of the AMS but of higher quality (or better taste?). For instance I have a PAMS paper and think that would be in the right ballpark of CR but that I would be lucky to get it published there.
It was suggested in the comments that the publications in CR are more like research announcements. I don't think that's true. They're short, punchy and written in a somewhat telegraphic style, but they certainly do include proofs. Quite recently I had the occasion to go to the actual library and pull off the shelf a CR paper. It was lovely, and short enough so that I transcribed (and translated, but big whoop: mathematical French is so easy that I can do it) it in its entirety on my notepad. (If you're interested, it is Guy Terjanian's first paper, referred to here...and as I learned slightly to my chagrin, one could regard the research contribution of this note of mine as being a fleshing out of a mild Alon-style generalization of Terjanian's argument. Actually there is another theorem he proves in that paper as well which is more interesting. To me this is a CR paper par excellence: a small but perfectly polished gem.)
I also think that Enseignement Mathematique is somewhere in this constellation of journals: more apparently elementary than CR, less laconic, but still high quality work which is somehow in "good taste". And note that the title of the journal would lead you to think that it publishes expository papers, but I don't think that's really the case.
After all this, let me come back to where I started:I am not sure what a "letter" is in this context, other than a short paper which is high quality and is written in a relatively laconic way. I have, unfortunately, zero expertise with physics, including the substantial portions of mathematics that overlap with physics. Maybe the concept of a letter is better understood by that portion of the mathematical community?
Added at the end: okay, let's see how MRL describes itself:
Dedicated to rapid publication of complete papers of original research in all areas of mathematics. Expository papers and research announcements of exceptional interest are also occasionally published. High standards are applied in evaluating submissions; the entire editorial board must approve the acceptance of any paper.
Thus there is some kind of vestigial connection with abbreviated papers, expository work and research announcements, but by and large it is no longer what that journal is about. I think this is rather typical.
Added: Since I was specifically asked to comment on PNAS, and I am a little gun-shy about leaving things in comments at the moment (see the meta site for more on this...), I will add the following non-answer answer: I have very little direct experience with PNAS. I tried to think of a single paper that was published in that journal and I came up with Milnor's "Eigenvalues of the Laplace operator on certain manifolds", a famous one page note. Based on that one paper (!!) I will guess that PNAS is like CR but for laconic treatments of even more important results. (I will also guess that most of their papers are more than one page long...)