Anki uses SQLite’s ‘like’ and ‘collate nocase’ features, which are fast, but not Unicode-aware, so they only work with the Latin alphabet. For searches, using a re: search will opt-in to the slower Unicode case matching.
A separate issue, but in the same vein: in the sidebar, searches are case-insensitive when using the Sidebar Filter, but when it comes to sorting, Saved Searches are sorted in a case-sensitive way (all the capitalized names come first), while Note Types and Tags are sorted in a case-insensitive way. (And other sidebar sections like Decks have a predefined or user-defined order rather than alphabetical.)