The character limits behind the compose counter used to be maintained in three separate places that could quietly disagree. They all read from one table now, with a test that fails if any copy ever drifts again.
You should not see any difference day to day — same limits, same warnings. The visible change: over-limit messages now name the platform the way the rest of the app does ("over telegram's 1024-character limit" rather than inconsistent capitalization), and a note that needs media on a media-only platform is rejected up front instead of failing at publish time once such platforms exist.