Why is Base91 denser than Base64?

Base64 always spends 4 output characters per 3 input bytes, a fixed 33 percent overhead. basE91 instead streams bits and emits either 13 or 14 bits per pair of characters depending on the value, so it averages about 14 percent overhead — meaningfully less for large payloads.

How does the variable bit width work?

The encoder fills a bit accumulator and, once it holds more than 13 bits, takes a 13-bit chunk when that value is greater than 88, otherwise a 14-bit chunk. Each chunk becomes two characters via value mod 91 and value over 91. The decoder mirrors this, reading 13 or 14 bits based on whether the reconstructed value exceeds 88.

What characters does Base91 use?

It uses 91 printable ASCII characters: all letters and digits plus a set of punctuation marks, while excluding the space, the apostrophe, the hyphen, and the backslash. Avoiding those keeps the output usable in many text contexts, though it is not specifically URL-safe.

Is Base91 a recognised standard?

It is a well-known scheme published by Joachim Henke rather than an IETF RFC. Implementations are widely compatible because they follow his reference algorithm, which is the exact bit-packing logic this tool implements.

Does my data leave the browser?

No. Encoding and decoding run entirely in JavaScript in your browser, so any binary or text you process stays on your device.

Base91 Encoder/Decoder

basE91 is a binary-to-text encoding by Joachim Henke designed for maximum density among printable schemes. By packing a variable 13 or 14 bits into each character pair, it carries roughly 14 percent overhead — noticeably tighter than Base64’s fixed 33 percent. This tool encodes UTF-8 text into Base91 and decodes it back, entirely in your browser.

How it works

Base91 streams bits through an accumulator rather than using fixed groups:

Input bytes are shifted into a bit accumulator, eight bits at a time.
Once the accumulator holds more than 13 bits, the low 13 bits are examined. If that value is greater than 88 the encoder consumes 13 bits; otherwise it consumes 14 bits. This keeps every two-character unit within the 8281 combinations that 91 squared provides.
The consumed value is split into two characters: value mod 91 and value / 91, each mapped through the 91-symbol alphabet.
Any leftover bits at the end are flushed as one or two final characters.

Decoding reverses the process, reading two characters into a value, then refilling and draining the bit accumulator a byte at a time, choosing 13 or 14 bits using the same greater-than-88 test.

Tips and example

Encoding the text test produces fPNKd — shorter than the six characters Base64 would use for the same four bytes. The density advantage grows with payload size, so Base91 is a good fit when you are squeezing binary into a text channel and every byte counts. It is not URL-safe out of the box, however: the alphabet includes characters like & and + that need escaping in query strings, so prefer Base62 when the result must live in a URL.