The docs for the OCR API say it should, but in practice it struggles with characters that aren’t latin alphabet letters, cyrillic letters, or numbers.