Two alphabets
An SMS carries 140 bytes. With the GSM-7 alphabet, seven bits per character, that is 160 characters. With Unicode (UCS-2), sixteen bits per character, that is 70. The alphabet is decided for the whole message: one character outside GSM-7 and every character is billed at the Unicode size.
You do not choose the encoding; the text does. The composer shows which one applies, with the segment count and the longest rendered row, and names the character that forced Unicode when there is one. The API never rewrites text: it returns the encoding, gsm7 or ucs2, and the number of segments with every message, as described in the message object.
What GSM-7 contains
The GSM-7 basic table has 128 positions: one escape code, two control codes for line breaks and 125 printable characters.
| Group | Characters |
|---|---|
| Letters | A to Z and a to z |
| Digits | 0 to 9 |
| Punctuation and symbols | space ! " # % & ' ( ) * + , - . / : ; < = > ? @ _ § ¡ ¿ ¤ |
| Currency | £, ¥ and the dollar sign; the euro sign is in the extension table |
| Accented and other Latin letters | à è é ì ò ù É Ç Ä Ö Ü ä ö ü ß Ñ ñ Å å Æ æ Ø ø |
| Greek capitals | Δ Φ Γ Λ Ω Π Ψ Σ Θ Ξ |
Line breaks are in the set, so a message on two lines stays in GSM-7, though each break uses a character. Notice what is missing: the backtick, the degree sign, curly quotes, the en and em dashes, the ellipsis character, the bullet, ©, ® and ™, and most accented letters beyond Western Europe. The Greek capitals are a historical leftover: Greek as it is normally written, with lower case and accents, needs Unicode.
The extension table
Nine more characters are reached through the escape code, so each one takes two positions: ^ { } \ [ ] ~ | €. They keep a message in GSM-7 but count as two characters each; the glossary lists them under GSM extended characters. A message of 159 visible characters with one euro sign counts 160 and fits in one segment; add a second euro sign and it counts 162, which is two.
They hide in prices, references and templates. A price written as €25 counts four characters, and square brackets around an order number cost four positions where parentheses cost two.
Segments and concatenation
A longer text is split into segments that the handset reassembles. Each segment gives up six bytes to a header that says "part 2 of 3", which is why the per-part capacity drops to 153 and 67. You pay per segment, up to 6. The composer and the API response both show the segment count before anything is sent, so the price is never a surprise.
| Segments | GSM-7 characters | Unicode characters |
|---|---|---|
| 1 | up to 160 | up to 70 |
| 2 | 161 to 306 | 71 to 134 |
| 3 | 307 to 459 | 135 to 201 |
| 4 | 460 to 612 | 202 to 268 |
| 5 | 613 to 765 | 269 to 335 |
| 6 | 766 to 918 | 336 to 402 |
Two things follow. The step from one segment to two is the expensive one: the 161st character buys a whole second segment, and the first part shrinks from 160 characters to 153. And six parts is the ceiling, so the longest message is 918 GSM-7 characters or 402 in Unicode. Near a boundary, trust the counter over the table: systems generally avoid splitting an escape sequence or an emoji across two parts, so a part can hold one character fewer than the arithmetic suggests.
Worked examples
One apostrophe, three segments. This reminder is exactly 160 characters:
ACME Dental: a reminder of your check-up on Tuesday at 14:30 with Dr Patel. Please arrive ten minutes early and bring your card. Can't come? Call 020 7946 0000.
Typed with a straight apostrophe, it is one GSM-7 segment. Pasted from a word processor that turned Can't into Can’t, it becomes 160 Unicode characters, and 160 falls in the 135 to 201 band: three segments, three times the price.
The longest name decides. A loyalty message whose fixed text is 156 characters, around a {first_name} placeholder:
ACME: Hi {first_name}, your loyalty card now holds 1,250 points, enough for a free coffee and a pastry. Show this message at the till in any of our shops before Sunday.
For Emma it renders at exactly 160 characters: one segment. For James it is 161: two segments. For Zoë it is 159 characters, but ë is not in GSM-7, so her message goes out in Unicode: three segments. Each message is encoded and billed on its own, so only those rows cost more, and the composer shows the longest rendered row and names any character that forced Unicode, so both problems are visible before you send.
What it costs. At an illustrative rate of USD 0.01 per segment, 10,000 messages in one segment cost USD 100. The same 10,000 pushed into three Unicode segments by one apostrophe cost USD 300. Each segment is billed at the destination's published rate, listed on the pricing page.
What switches the encoding
GSM-7 covers the basic Latin letters, digits, common punctuation and a handful of accented characters used in Western European languages. It does not cover smart quotes, the em dash, emoji, or letters like ő, ș, ă. Any one of those switches the message to Unicode. The most common accidental switch is a curly apostrophe pasted from a word processor; phone keyboards with smart punctuation turned on produce the same characters.
| Character | Usually comes from | Type instead |
|---|---|---|
| ’ ‘ curly apostrophe and single quotes | Word processors, smart punctuation | ' |
| “ ” curly double quotes | Word processors, smart punctuation | " |
| – — en and em dashes | Autocorrect of a double hyphen | - |
| … ellipsis character | Autocorrect of three dots | ... |
| Non-breaking and other special spaces | Text copied from web pages and documents | An ordinary space |
| • bullet | Copied lists | A hyphen or a line break |
| ° degree sign | Temperatures | The word: 21 degrees |
| ™ ® © | Brand names | Nothing |
| ` backtick | Code, some keyboard layouts | ' |
| Zero-width and other invisible characters | Copied text, some content systems | Retype the line |
| Emoji | Phone keyboards | A word |
The composer offers one-click replacements for curly quotes, long dashes, the ellipsis character and special spaces. The API sends text exactly as submitted, so a script that takes copy from a content system should apply the same replacements before it sends.
Emoji need a word of their own, and the glossary has more on emoji in SMS. They are never GSM-7, and most also sit outside the range UCS-2 was designed for, so they travel as a pair of 16-bit units: one emoji usually takes two of the 70 places. Skin tones, flags and combined emoji take four or more.
Accents, language by language
GSM-7 was built around Western European languages and covers them unevenly. One letter from the right-hand column is enough to switch a message:
| Language | In GSM-7 | Switches to Unicode |
|---|---|---|
| English | Everything in ordinary text | Only typographic punctuation |
| German | ä ö ü Ä Ö Ü ß | Nothing in everyday text |
| French | à é è ù É | â ê î ô û ë ï œ, and capitals such as À and È |
| Spanish | é ñ Ñ ü ¿ ¡ | á í ó ú and their capitals |
| Portuguese | à é É | á â ã ê í ó ô õ ú |
| Italian | à è é ì ò ù | È at the start of a sentence, and other accented capitals except É |
| Danish, Finnish, Norwegian, Swedish | å ä æ ö ø and their capitals | Nothing in everyday text |
| Central and Eastern European, Turkish | é, ö and ü where the language uses them | Most other diacritics: ą ę ł ś ż, č ř š ž, ő ű, ă ș ț, ğ ı ş |
The c with cedilla is a special case: the standard table holds a capital Ç, and systems disagree on whether a lower-case ç maps to it, so French, Portuguese and Turkish text that uses it can go either way. Check the counter rather than assuming. Beyond that, the choice is editorial. Spanish or Portuguese without accents stays in GSM-7 and costs less, but some audiences read it as careless. For a one-time code it rarely matters; for a brand campaign, decide on purpose.
Writing for one segment
To check a draft as you write, paste it into the SMS character counter: it names each character that forces Unicode and shows the segment count before and after replacing it.
- Type straight quotes and a plain hyphen.
- Skip emoji in transactional messages: they cut the capacity from 160 characters to 70 for no delivery benefit.
- Keep links short and stable; a tracking link of 30 characters is a fifth of the segment.
- Preview with real data: a variable like a name can push the longest row over the limit while the template fits.
- Prefer parentheses to square brackets, and 14:30 to 2:30 p.m.
- Cut words before you cut clarity: Your order has shipped says what We are pleased to let you know that your order is now on its way says, in a third of the space.
The blog post on Unicode cost and how to avoid it covers the cost side in more detail.
Non-Latin scripts
Arabic, Cyrillic, Hebrew, Devanagari and the other Indic scripts, Thai, Chinese, Japanese and Korean are always Unicode, and so is Greek as it is normally written. That is not a problem to solve, it is the cost of the script: 70 characters per single message, 67 per part. A Latin brand name inside such a message changes nothing, because the message is already Unicode.
Codes, alerts and short confirmations fit in one segment in any language. ACME: ваш код 493201. Действует 10 минут. is 41 characters, comfortably one segment; the guide to verification codes covers the rest of the format. Write marketing copy in two parts on purpose rather than in one and a half by accident. In Chinese and Japanese, where a single character often carries a whole word or syllable, 70 characters go further than they look. Some senders transliterate, writing Russian or Greek in Latin letters to stay in GSM-7; it more than doubles the room per segment but reads as informal, which suits a code better than a brand.
Something unclear or out of date? Tell us on the support page; corrections are folded back into the guide and the update date changes.
