The 'fmt' tool uses bytes to count line length instead of measuring how wide (in columns) the text actually is. This yields strange results when working with multibyte encodings, such as UTF-8. Note the manual page also uses the term "columns", not "bytes" or "characters". If you implement this feature, remember some characters, e.g. CJK, can take more than one column, too.
Well - actually none of the coreutils utilities upstream support unicode characters handling. We have nasty&ugly downstream patch in Fedora to cover some of the utilities and I don't plan to extend it. Marking as future feature, as Ondrej Oprala works on the support upstream, so it might be done eventually - but not as a part of the current downstream patch.
This package has changed ownership in the Fedora Package Database. Reassigning to the new owner of this component.
Ondrej Oprala does not work on this any more. It is unlikely that Fedora maintainers could make any progress on this in the near future. Closing WONTFIX.