UCS-2 is the 16-bit encoding used when a message contains characters outside the GSM-7 alphabet; it allows 70 characters per message (67 per part), so non-Latin text costs about twice as many segments.
When a text contains any character the GSM-7 tables do not have, the handset and network switch the entire message to UCS-2, where every character takes two bytes. The 140-byte payload then holds 70 characters, or 67 per part in a concatenated message.
Scripts such as Cyrillic, Arabic, Hebrew, Devanagari, Thai, Chinese, Japanese and Korean are always UCS-2; there is no cheaper way to send them and campaigns to those countries should be planned in 70-character units. For Latin text, UCS-2 is almost always accidental: a pasted apostrophe, a bullet, an emoji.
At SMSMeteor: Unicode messages hold 70 characters (67 per part when concatenated); the composer shows the offending character and suggests a GSM-7 replacement when one exists.