What's your field? In my field (math), I am not aware of any journals that publish an HTML version.

All of Springers math journals publishes every new paper in a web-version (using HTML and related technologies). Here is a random example: link.springer.com/article/10.1007/s13373-016-0093-2/…

@NateEldredge There are plenty but I recommend avoiding the HTML version at all costs. The conversion process often introduces errors and I recently found a paper where these were significant (IIRC to the point of making theorems obviously false).

@DavidRicherby: that’s interesting — do you remember what the particularly bad example was? It would be good to see what kind of specific errors to be on guard for.

@DavidRicherby I wouldn't. The HTML version has several advantages which make them easier to read. The mentioned Springer HTML version offenr: The citations open the details of the citation including links to MathSciNet, Google Scholar and others, jumping to footnotes and back is easy as well as following links to sections or equations and jumping back. Regarding errors: I would assume that the pdf and the html version undergo the same editing, so I'd guess that the pdf would also suffer from errors, but I am not sure about this.

@Dirk For what it's worth, all the hyperlinking you mention can be done in PDF as well (and easily through LaTeX), but without all the ugliness of trying to "typeset" mathematics in HTML.

@jeffE Sure, but the jump-back does not work with every pdf-viewer and the pop-up functionality is missing in pdf (as far as I know). Regarding ugliness, I am actually pretty happy with the HTML-versions I have read. Another upside is that you can actually extract and search the TeX-code behind the math. (A related downside is that you sometime see some weird TeX-code…)

re "I would assume that the pdf and the html version undergo the same editing": I've always been asked to check the proof of the PDF version of my paper in the editing phase, but I have never even seen the HTML version before it appears on the journal's website. I wouldn't be surprised if the HTML version contains errors that are not in the PDF. (And, for example, some colleagues have noticed errors in the PDF version that can only be the result of the copy-editor copying numbers by hand to the final tables, so making the HTML version from the might contain similar nontrivial steps.)

HTML is not computer-friendly per se, at least not until something like MathML becomes standard. (MathJax is a good step forward for HTML, but it's less suited for reading than PDF. For example, try taking a MathJax website with you on a tablet to a place with no internet. Or just reading one with a slow connection. Actually, MathOverflow doesn't work very well on my Android tablet even with a perfectly fine connection.)

@user60974: Thanks, I had not seen that before. However it apparently isn't every Springer math journal, e.g. link.springer.com/journal/11118 (which I checked before posting) doesn't do it, unless I have overlooked something.

HTML is a step forward, however it does not work very well off-line,, because images are not embedded in the file and need to be saved in a separate folder.

@darij grinberg: MathJax (at least as used on StackExchange) is not much of a step forwards. Displaying black text when the user's background color is black doesn't really produce readable output, you know - though it's surprising how many web developers can't figure this out. Whereas pdf files are readable on just about any device, with or without internet connection.

IEEE does this, too, for recently published papers.

@NateEldredge Neither am I, and I am someone who would much prefer an HTML version due to a disability. See my comment to 1006a's answer below.

Related: tex.stackexchange.com/questions/300804/…

Some publishers do, e.g. Math Science Publisher provides two pdf versions of each paper, one for screen (with color hyperlinks etc.) One optimized for printing.

@PLL I dug out the example. see the discussion after Theorem 25 of On Treewidth Approximations (Bouchitté et all, Disc. Appl. Math., 136(2-3):183-196, 2004). The HTML version says "ρ′ is such that 12<ρ′<1. By taking... ρ′=(1+6)/2". The first expression gives an impossible requirement; the second looks like a really weird way of writing 7/2. But, the PDF says "ρ′ is such that ½<ρ′<1. By taking... ρ′=(1+√6)/2".

@Dirk See my comment immediately above this one. Changing mathematical expressions into nonsense is a total deal breaker.

@DavidRicherby Wow, that's really a mess. Did you contact the authors so that they can complain to Elsevier or did you send your complaint directly to Elsevier?