Universal character encoding detector.

chardet 7 is a ground-up, 0BSD-licensed rewrite of chardet. Same package name, same public API — drop-in replacement for chardet 5.x/6.x, just much faster and more accurate. Python 3.10+, zero runtime dependencies, works on PyPy.

Read more details about the rewrite process.

Why chardet 7?

99.7% accuracy on 3,138 test files. 315x faster than chardet 6.0.0, and +13.1pp more accurate than charset-normalizer 3.5.1 while being 1.2x faster. Language detection for every result. MIME type detection for binary files. 0BSD licensed.

Large inputs don't need a byte cap: detect(data, max_bytes=len(data)) on a 272 MiB file finishes in about 0.13-0.23s, and the UTF-8 verdict is validated over every byte examined, never sampled. See the performance docs for the full large-input tables.

Installation

pip install chardet

Quick Start

import chardet
chardet.detect(b"Python is a great programming language for beginners and experts alike.")
# {'encoding': 'ascii', 'confidence': 1.0, 'language': 'en', 'mime_type': 'text/plain'}
# UTF-8 English with accented characters
chardet.detect("The naïve approach doesn't always work in complex systems.".encode("utf-8"))
# {'encoding': 'utf-8', 'confidence': 0.84, 'language': 'en', 'mime_type': 'text/plain'}
# Japanese EUC-JP
chardet.detect("日本語の文字コード検出テストです。このテキストはEUC-JPでエンコードされています。正しく検出できるか確認します。".encode("euc-jp"))
# {'encoding': 'EUC-JP', 'confidence': 1.0, 'language': 'ja', 'mime_type': 'text/plain'}
# Get all candidate encodings ranked by confidence
text = "Le café est une boisson très populaire en France et dans le monde entier."
results = chardet.detect_all(text.encode("windows-1252"))
for r in results[:4]:
    print(r["encoding"], round(r["confidence"], 2))
# Windows-1252 0.31
# iso8859-15 0.31
# ISO-8859-1 0.31
# MacRoman 0.3

Streaming Detection

For large files or network streams, use UniversalDetector to feed data incrementally:

from chardet import UniversalDetector
detector = UniversalDetector()
with open("unknown.txt", "rb") as f:
    for line in f:
        detector.feed(line)
        if detector.done:
            break
result = detector.close()
print(result)

Encoding Era Filtering

Restrict detection to specific encoding eras to reduce false positives:

from chardet import detect_all
from chardet.enums import EncodingEra
data = "Москва является столицей Российской Федерации и крупнейшим городом страны.".encode("windows-1251")
# All encoding eras are considered by default — 4 candidates across eras
for r in detect_all(data):
    print(r["encoding"], round(r["confidence"], 2))
# Windows-1251 0.46
# MacCyrillic 0.42
# KZ1048 0.2
# ptcp154 0.2
# Restrict to modern web encodings — 1 confident result
for r in detect_all(data, encoding_era=EncodingEra.MODERN_WEB):
    print(r["encoding"], round(r["confidence"], 2))
# Windows-1251 0.46

Encoding Filters

Restrict detection to specific encodings, or exclude encodings you don't want:

# Only consider UTF-8 and Windows-1252
chardet.detect(data, include_encodings=["utf-8", "windows-1252"])
# Consider everything except EBCDIC
chardet.detect(data, exclude_encodings=["cp037", "cp500"])

CLI

chardetect somefile.txt
# somefile.txt: utf-8 with confidence 0.99
chardetect --minimal somefile.txt
# utf-8
# Include detected language
chardetect -l somefile.txt
# somefile.txt: utf-8 en (English) with confidence 0.99
# Only consider specific encodings
chardetect -i utf-8,windows-1252 somefile.txt
# somefile.txt: utf-8 with confidence 0.99
# Pipe from stdin
cat somefile.txt | chardetect
# stdin: utf-8 with confidence 0.99

What's New in chardet 7?

• 0BSD license (previous versions were LGPL)
• Ground-up rewrite: 13-stage detection pipeline using BOM detection, magic number identification, structural probing, byte validity filtering, and bigram statistical models
• 315x faster than chardet 6.0.0 when compiled, and 1.2x faster than charset-normalizer 3.5.1 overall (2.0x at the median) while being far more accurate
• 99.7% accuracy: +15.3pp vs chardet 6.0.0, +13.1pp vs charset-normalizer 3.5.1
• Language detection: 91.8% accuracy across 49 languages, returned with every result
• MIME type detection: identifies 40+ binary file formats (images, audio/video, archives, documents, executables, fonts) via magic number signatures, plus text/html, text/xml, and text/x-python for markup
• Encoding filters: include_encodings and exclude_encodings parameters to restrict or exclude specific encodings from the candidate set
• 99 encodings: full coverage including EBCDIC, Mac, DOS, and Baltic/Central European families
• Optional compiled builds: mypyc plus a Cython scoring kernel, 4.8x additional speedup on CPython
• Thread-safe: detect() and detect_all() are safe to call concurrently; scales on free-threaded Python
• Same API: detect(), detect_all(), UniversalDetector, and the chardetect CLI all work as before

Documentation

Full documentation is available at chardet.readthedocs.io.

Project History

chardet was originally created by Mark Pilgrim in 2006 as a Python port of Mozilla's universal charset detection library. He released versions 1.0 (2006) and 1.0.1 (2008) on PyPI, then developed an unreleased Python 3 port (2.0.1) on Google Code. After Mark deleted his online accounts in 2011, the project was continued by David Cramer, Erik Rose, Toshio Kuratomi, Ian Cordasco, and Dan Blanchard.

In 2026, Dan Blanchard rewrote chardet using Claude, releasing chardet 7.0 under a new license. All releases after 7 are not derivative of the original chardet code, but are released under the same name to allow an easier transition for users who can immediately benefit from the speed and accuracy improvements. For historical preservation and to allow easier comparison with the other releases, Dan has restored Mark's lost commits to this repository in the history/pilgrim branch.

To see the full history from 2006 to present in git log, fetch the graft refs:

git fetch origin 'refs/replace/*:refs/replace/*'

License

0BSD