Skip to the article
On this page
  1. The four common data modes
  2. Why numeric mode fits more digits
  3. What happens when the message contains lowercase letters?
  4. Software test: how the same content fits into different modes
  5. A QR can mix modes
  6. Kanji mode and Unicode are not synonyms
  7. Can one long message span more than one QR code?
  8. Does a mode determine security?
  9. Frequently Asked Questions
  10. What to check before publishing a QR code
  11. Sources and further reading

QR Code Encoding Modes: Numeric, Alphanumeric, Byte, and Kanji

By OwnQR · Published · Updated

QR encoding modes determine how a segment of data is represented as bits. The four commonly used modes are numeric, alphanumeric, byte and Shift_JIS-based Kanji. A QR generator normally chooses the appropriate representation for ordinary website URLs; mode choice helps explain why similar-looking text can occupy different amounts of space. These are encoding techniques, not four separate consumer QR products.

That distinction is helpful when a QR suddenly becomes denser after adding punctuation, lowercase letters, or non-Latin text. It also explains why the maximum capacity figures in product brochures are not interchangeable.

The four common data modes

DENSO WAVE lists four conventional data categories in its QR outline specification:

ModeWhat it is designed to representExample concept
NumericDigits 0–9 efficientlyAn all-digit identifier
AlphanumericSupported uppercase letters, digits and selected symbolsA restricted uppercase reference
ByteArbitrary byte values, interpreted through an agreed character encodingMixed-case URL or multilingual text bytes
KanjiSpecific eligible characters represented through the Shift_JIS character encodingCompatible Japanese text

A mode is a bitstream representation. An app's label “Wi-Fi QR,” “payment QR,” or “social media QR” describes the meaning of the payload, not necessarily the mode that the QR encoder chooses to represent it.

A QR symbol format and an encoded data mode answer different questions. Models such as Micro QR and rMQR concern the shape and arrangement of the symbol; numeric, alphanumeric, byte and Kanji concern the way data is packed. The types of QR codes guide separates those categories. For visible reserved modules and masking, see QR code anatomy.

What happens to accented characters?

Accented letters such as é are not members of QR Code's restricted alphanumeric character set. They can instead be represented as bytes using an agreed character encoding; for example, UTF-8 represents é as two bytes (C3 A9), whereas the plain ASCII letter e is one byte. That distinction can affect payload size, but does not by itself determine the QR version. The Unicode UTF-8 specification defines UTF-8 encoding, while DENSO WAVE's QR specification overview identifies the byte data mode. The reader must interpret the bytes using the intended encoding; do not assume every scanner handles non-ASCII text identically without testing.

Why numeric mode fits more digits

Restricted alphabets let an encoder represent information more compactly. In a standard Model 2 QR data segment, a 4-bit mode indicator identifies the representation. A character-count field, whose width depends on the mode and QR version, tells the decoder how many characters or bytes follow. The data bits then encode that segment. Segments can be mixed within a symbol, so capacity depends on more than the character count.

ModeTypical data packing within a segment
NumericThree digits per 10 bits; a remaining pair uses 7 bits, or one digit 4 bits
AlphanumericTwo characters per 11 bits; one remaining character uses 6 bits
Byte8 bits per byte (one character may require several bytes)
Kanji13 bits per eligible Shift_JIS character

These values describe data-segment packing, not the complete QR symbol. In particular, the 7-, 4-, and 6-bit values are for a final short group of input characters; they are not the separate, version-dependent remainder bits added during QR construction. For the trailing numeric and alphanumeric groups, see the worked numeric mode explanation and alphanumeric mode explanation. Additional implementation references and the DENSO WAVE outline specification are listed under Sources.

For example, the 20-digit string in the test below contains six groups of three digits plus a final pair. Numeric mode needs 6 × 10 + 7 = 67 data bits for those digits; byte mode needs 20 × 8 = 160 data bits for the same ASCII digits.

For a separate three-character alphanumeric example, one 11-bit pair plus a 6-bit final character requires 17 data bits, versus 24 data bits for three ASCII bytes in byte mode. Both examples are calculations from the published packing rules. They exclude segment headers and error correction and are not phone or printing experiments.

Separate arithmetic cross-check (our own): A software calculation applied the cited grouping rules to 160 entries in GS1 General Specifications Release 26.0, Table 5-36 (Versions 1–10 × four error-correction levels × four modes). The printed table lists 24 for the Version 7, level-M data-codeword entry; this cross-check used 124 for that entry, consistent with the arithmetic, rather than silently claiming to reproduce the printed value as-is. This is our software comparison—not an independent third-party audit, a scan-compatibility study, or a measurement of OwnQR's production generator.

The famous maximum figures—for example, up to 7,089 numeric characters and 2,953 8-bit bytes under the standard's highest-capacity configurations—should not be used as promises for arbitrary real-world messages. Version selection, error-correction level, mixed segments and other constraints change what fits.

What happens when the message contains lowercase letters?

Standard QR alphanumeric mode supports an explicit restricted set of characters; ordinary lowercase letters are not in that set. An encoder can switch modes or use byte data, which can increase the amount of space required. Two encoders may choose different segmentations for the same input and still produce valid symbols.

Consider an uppercase reference code versus a URL containing lowercase paths and query parameters. The second may require byte segments or a mixed strategy. A longer, denser pattern is not automatically a defect. It reflects the data and the encoder's decisions.

How do I keep a dense code readable on a small label?

Try these changes in order:

  1. Shorten the payload. For a web link, remove unnecessary parameters and use a short, stable URL. Shorter content often allows fewer modules. DENSO WAVE's version and capacity guide explains how capacity and version relate.
  2. Use uppercase only when it preserves meaning. An identifier you control can use QR alphanumeric mode when every character belongs to its set: digits 0–9, uppercase A–Z, space, and $ % * + - . / :. Lowercase letters are excluded. A URL's scheme and host are case-insensitive, but its path and query may be case-sensitive; do not uppercase an existing URL without verifying the destination. See RFC 3986 §6.2.2.1.
  3. Compare error correction at L, M, Q and H. Lower levels may allow a smaller symbol at the cost of less recovery from damage; H has greater damage tolerance but may need more modules. Keep a four-module quiet zone around a standard QR Code and test the final print at the actual label size. See DENSO WAVE's error-correction explanation and printing-area guidance.
  4. Consider Micro QR only for a very short payload and a known compatible reader. DENSO WAVE's Micro QR specification lists M4 as the largest Micro QR symbol (17 × 17 modules), with maxima of 35 numeric characters, 21 alphanumeric characters at M4-L, or 15 bytes at M4-L. That byte limit rules out most complete URLs. Short numeric or uppercase identifiers are better candidates. Smartphone support was not tested here, so do not assume a public phone camera can read Micro QR.

Measured software-only comparison. In a first-party local test dated November 16, 2026, Claude generated clean QRCodeSVG images with qrcode.react 4.2.0 (zero encoder margin), then rendered them at eight pixels per module with an added four-module quiet zone. Both jsQR 1.4.0 and zxing-cpp via zxing-wasm 3.1.7 decoded every image back to its exact payload. These are recorded results from this encoder and two software decoders, not a phone, physical print, minimum-size or OwnQR production-generator test.

Exact payloadError correctionObserved versionSymbol modulesjsQRzxing-cpp
https://example.com/product/ABC-12345 (37 characters)LV329 × 29PassPass
Same 37-character payloadMV329 × 29PassPass
Same 37-character payloadQV433 × 33PassPass
Same 37-character payloadHV537 × 37PassPass
https://example.com/ followed by 100 lowercase a characters (120 characters)LV641 × 41PassPass
Same 120-character payloadMV745 × 45PassPass
Same 120-character payloadQV953 × 53PassPass
Same 120-character payloadHV1161 × 61PassPass

The test generated eight images (two payloads × four error-correction levels); both decoders passed all eight images, for 16 successful decoder checks. For the 37-character payload, L-to-H increased the symbol width from 29 to 37 modules (about 28%); for the 120-character payload, it increased from 41 to 61 (about 49%). At a fixed printed width, more modules mean smaller individual cells. These results do not establish that any of these symbols will fail to scan: all decoded successfully in the clean software test.

Physical dimensions also include the margin. DENSO WAVE's worked example uses 29 modules at 0.254 mm per module, yielding 7.366 mm for the symbol itself; adding the standard four-module quiet zone on each side makes 37 modules total, or 9.398 mm at that module pitch. This is a dimensional example, not a measured phone-scanning threshold.

Software test: how the same content fits into different modes

To make the encoding difference concrete, a software-only test used segno 1.6.6 for encoding and zxing-cpp 2.3.0 for decoding at error-correction level M. These are synthetic examples, not claims about how OwnQR itself selects modes or how any phone-camera app behaves. All listed samples decoded back to their original text in this setup.

Exact input or testSelected / forced modeObserved QR version
01234567890123456789 (20 digits)Numeric, automaticV1 (21 × 21)
Same 20 digits, forced byteByteV2 (25 × 25)
HTTPS://EXAMPLE.COM/MENUAlphanumericV2 (25 × 25)
https://example.com/menuByteV2 (25 × 25)
日本語テキストKanji (Shift_JIS)V1 (21 × 21)
繁體中文菜單Kanji (Shift_JIS)V1 (21 × 21)
Menu 🍜🍣Byte (UTF-8, no ECI; encoder default)V1 (21 × 21)
繁體中文菜單 🍜 with explicit UTF-8 + ECIByteV2 (25 × 25)

The emoji example used the encoder’s default UTF-8 byte interpretation without an ECI designator. Its decoding in this one test does not establish that every scanner will interpret the bytes identically. The forced-byte digit example grew from V1 to V2. But not every switch to byte mode increases the version: the two example URLs both selected V2. The result depends on payload length, segmentation, character encoding and error correction, not merely whether a URL contains lowercase text.

A QR can mix modes

A QR bitstream can contain several segments, each with mode information and an appropriate count. An encoder may represent different parts of a message in different modes to avoid wasting capacity. The standard also provides mechanisms such as Extended Channel Interpretation (ECI) for character-set handling, but a compatible reader must understand the selected interpretation.

This means it is misleading to classify every finished QR image as one immutable “type.” A single symbol's payload can be encoded using a mixture of techniques. The visible symbol family—for example, standard QR versus Micro QR—is yet another separate classification.

What does a reader assume when there is no ECI?

Without an ECI header, the ISO/IEC 18004:2005 byte-mode default is ISO-8859-1; the 2000 edition used JIS8, according to an answer quoting those editions, rather than the paywalled standard text directly. ECI (Extended Channel Interpretation) explicitly signals how subsequent bytes should be interpreted: ECI 000003 identifies ISO-8859-1 and ECI 000026 identifies UTF-8, as defined in ZXing's character-set mapping. The ECI designator identifies the interpretation for subsequent byte data; it is not an encryption mechanism.

In practice, unmarked UTF-8 can still decode correctly on some readers. For example, ZXing's QR decoder uses character-set guessing when no ECI is supplied, and its detection logic considers UTF-8, ISO-8859-1 and Shift_JIS; Thonky's byte-mode guide also notes that some readers recognize UTF-8 without ECI. An ECI designator requires additional header bits and can increase symbol size near a capacity boundary, while readers that ignore or mishandle ECI may still display text incorrectly. Prefer explicit UTF-8 ECI when strict interpretation matters and your readers support it; where capacity is tight and readers are known to infer UTF-8, omitting ECI may be a compatibility choice. Test the exact decoded multilingual text on the intended readers rather than assuming either approach works universally.

Kanji mode and Unicode are not synonyms

Kanji mode is a specialized compaction method for characters supported by the relevant Shift_JIS encoding, not a universal Unicode shortcut. In one software example, the Han characters in 繁體中文菜單 were eligible for this mode, and the encoder chose Kanji (Shift_JIS). That does not mean all Traditional Chinese characters, Korean text, or emoji qualify. A decoder must interpret the character encoding correctly; for broader Unicode content, UTF-8 byte data with appropriate character-set signaling such as ECI may be used, where supported by the reader.

For emoji and other multilingual strings, distinguish characters from UTF-8 bytes. A single displayed character can occupy several bytes, so byte-mode capacity depends on bytes, not on how many symbols a person sees on screen.

For a URL, non-ASCII text may need to be represented as UTF-8 bytes and then percent-encoded when creating the URI. That is URL representation, not QR encryption; RFC 3986 describes URI percent-encoding.

Finally, test interoperability as a separate task. Compare the actual decoded text or bytes with the intended input and, when possible, check how the receiving app handles URLs, contacts or Wi-Fi configurations. A correctly encoded QR symbol and a working app action are different outcomes.

Can one long message span more than one QR code?

The QR standard includes structured append, allowing a message to be split across related symbols; both DENSO WAVE's standards overview and GS1 General Specifications Release 26.0 (§5.7.3) describe a maximum of 16 symbols for this feature. Do not assume an ordinary phone-camera app automatically combines them. In one software-only test, segno was instructed to use Version 5 and generated six symbols from a 420-character synthetic URL. zxing-cpp decoded each symbol separately, returning 70 characters for each individual symbol. The test did not attempt to merge the results, so it cannot establish whether the decoder or a camera app would automatically reconstruct the full message. A multi-symbol workflow needs compatible software that can collect and reconstruct the full content.

Does a mode determine security?

No. Mode choice is not encryption. Standard numeric or byte payloads can normally be decoded by compatible readers; encryption, where needed, belongs to a separate system layer. Our guide Are QR codes encrypted? explains how encoding, encryption and HTTPS differ.

Frequently Asked Questions

Is byte mode the same as binary mode?

Byte mode carries byte values and is often referred to as an 8-bit or binary data mode. The interpretation of those bytes can depend on encoding information and the reader.

How are QR codes encoded?

An encoder converts the input into one or more data segments, writes a mode indicator and count field for each segment, packs the data bits, and constructs a QR symbol with error-correction and fixed function patterns. A compatible decoder reverses those steps. Multiple modes can appear within one QR.

Is Kanji mode only for Japanese?

Not strictly. Kanji mode represents eligible characters through a defined Shift_JIS-based mapping, so it can include Han characters used in Chinese when those exact characters are present in the supported Shift_JIS set. In our software example, 繁體中文菜單 was eligible. This does not mean all Traditional Chinese, Korean or other Unicode text qualifies; unsupported characters need other encoding, commonly byte mode.

How does UTF-8 text encoding work in a QR code?

UTF-8 represents Unicode text as bytes. It is a character encoding, rather than an additional QR data mode; QR byte-mode segments can carry encoded bytes. When estimating capacity, count encoded bytes rather than visible characters: for example, é takes two UTF-8 bytes and many emoji take four. See the Unicode Standard's UTF-8 chapter and RFC 3629.

For reliable multilingual decoding, check whether your encoder and intended readers support explicit character-set designation, such as ECI, and test the actual decoded text on those readers. Do not assume that every reader interprets unmarked bytes identically. This is a compatibility recommendation, not a claim that a particular scanner or OwnQR workflow has been tested.

What to check before publishing a QR code

  1. Measure byte length, not only visible characters. Emoji and non-Latin text may need more than one byte each.
  2. Check the actual encoder result. Mixed-mode decisions, error correction and a URL's characters can change the QR version.
  3. Decode the generated symbol and compare the result. Check the exact intended text or URL, especially for non-ASCII content and ECI handling.
  4. Keep payload capacity and scan reliability separate. A software decode does not prove a printed code will work on every phone.

For related reading, start with the QR Code basics guide for the overall process and the QR code anatomy guide for finder and timing patterns. For capacity data, consult DENSO WAVE's published QR specifications together with the actual encoder settings.

Sources and further reading

  • DENSO WAVE — QR Code outline specifications (publication date not stated; study/event period: not applicable; accessed 2026-11-10): data modes, capacity categories, and structured appending of up to 16 symbols.
  • DENSO WAVE — QR Code FAQ (publication date not stated; study/event period: not applicable; accessed 2026-11-08): data encoding and character-set considerations.
  • DENSO WAVE — What Is a QR Code? (publication date not stated; study/event period: not applicable; accessed 2026-11-08): QR features and general principles.
  • ISO/IEC 18004:2024 — QR Code symbology catalogue (published 2024-08; study/event period: not applicable; catalogue accessed November 2026): official scope, including data encoding methods. This article does not claim to reproduce the paid standard's complete technical text.
  • Thonky — QR Code Data Encoding (publication/update date: not independently verified; study/event period: not applicable; accessed 2026-11-08): mode indicators and character-count fields.
  • Thonky — Numeric Mode Encoding (publication/update date: not independently verified; study/event period: not applicable; accessed 2026-11-08): 10/7/4-bit grouping, including final short groups.
  • Thonky — Alphanumeric Mode Encoding (publication/update date: not independently verified; study/event period: not applicable; accessed 2026-11-08): 11-bit pairs and 6-bit remaining character.
  • PHP-QRCode technical terminology (publication date not stated; study/event period: not applicable; accessed 2026-11-08): ECI and mixed-mode mechanics; implementation documentation.
  • GS1, General Specifications, Release 26.0 (published January 2026; study/event: NOT_APPLICABLE; accessed November 10, 2026): §5.7.2 documents 8-bit byte and 13-bit Kanji representations, §5.7.3 references up to 16 structured-append symbols; Table 5-36 provides checked QR capacities.
  • RFC 3986 — URI Generic Syntax (published 2005-01; study/event period: not applicable; accessed November 10, 2026): URI percent-encoding of bytes.
  • Software-only editorial experiment for OwnQR (study/event period: one run no later than November 8, 2026): the open-source segno 1.6.6 encoder and zxing-cpp 2.3.0 decoder processed synthetic text; this was not the OwnQR production generator, and no phone, printing or real-user test was conducted.

Prepared with AI assistance. This article has not been independently reviewed by a human technical expert.

Browse all QR code guides