Sourced from chardet's releases.

7.4.1

Bug Fixes

• BOM-prefixed UTF-16/32 input now returns utf-16/utf-32 instead of utf-16-le/utf-16-be/utf-32-le/utf-32-be. The endian-specific codecs don't strip the BOM on decode, so callers were getting a stray U+FEFF at the start of their text. BOM-less detection is unchanged. (#364, #365)

Full Changelog: chardet/chardet@7.4.0...7.4.1

chardet 7.4.0 brings accuracy up to 99.3% (from 98.6% in 7.3.0) and significantly faster cold start thanks to a new dense model format.

What's New

Performance:

• New dense zlib-compressed model format (v2) drops cold start (import + first detect) from ~75ms to ~13ms with mypyc

Accuracy (98.6% → 99.3%):

• Eliminated train/test data overlap via content fingerprinting
• Added MADLAD-400 and Wikipedia as supplemental training sources
• Improved non-ASCII bigram scoring: high-byte bigrams are now preserved during training and weighted by per-bigram IDF
• Encoding-aware substitution filtering (substitutions only apply for characters the target encoding can't represent)
• Increased training samples from 15K to 25K per language/encoding pair

Bug fixes:

• Added dedicated structural analyzers for CP932, CP949, and Big5-HKSCS (these were previously sharing their base encoding's byte-range analyzer, missing extended ranges)

Metrics

	chardet 7.4.0 (mypyc)	chardet 6.0.0	charset-normalizer 3.4.6
Accuracy (2,517 files)	99.3%	88.2%	85.4%
Speed	551 files/s	12 files/s	376 files/s
Language detection	95.7%	40.0%	59.2%

Full changelog: https://chardet.readthedocs.io/en/latest/changelog.html

7.3.0

License

• 0BSD license — the project license has been changed from MIT to 0BSD, a maximally permissive license with no attribution requirement. All prior 7.x releases should also be considered 0BSD licensed as of this release.

Features

• Added mime_type field to detection results — identifies file types for both binary (via magic number matching) and text content. Returned in all detect(), detect_all(), and UniversalDetector results. (#350)
• New pipeline/magic.py module detects 40+ binary file formats including images, audio/video, archives, documents, executables, and fonts. ZIP-based formats (XLSX, DOCX, JAR, APK, EPUB, wheel, OpenDocument) are distinguished by entry filenames. (#350)

Bug Fixes

• Fixed incorrect equivalence between UTF-16-LE and UTF-16-BE in accuracy testing — these are distinct encodings with different byte order, not interchangeable

Performance

... (truncated)

Sourced from chardet's changelog.

7.4.1 (2026-04-07)

Bug Fixes:

• BOM-prefixed UTF-16 and UTF-32 input now reports utf-16 and utf-32 instead of the endian-specific variants. Python's utf-16-le/utf-16-be/utf-32-le/utf-32-be codecs keep the BOM as a U+FEFF in the decoded string, while utf-16/utf-32 strip it, so callers passing the detection result directly to .decode() were getting a stray BOM at the start of their text. BOM-less UTF-16/32 detection (via null-byte patterns) is unchanged and still returns the endian-specific name. (Dan Blanchard <https://github.com/dan-blanchard>_ via Claude, [#364](https://github.com/chardet/chardet/issues/364) <https://github.com/chardet/chardet/issues/364>, [#365](https://github.com/chardet/chardet/issues/365) <https://github.com/chardet/chardet/pull/365>)

7.4.0 (2026-03-26)

Performance:

• Switched to dense zlib-compressed model format (v2): models are now stored as contiguous memoryview slices of a single decompressed blob, eliminating per-model struct.unpack overhead. Cold start (import + first detect) dropped from ~75ms to ~13ms with mypyc. (Dan Blanchard <https://github.com/dan-blanchard>_ via Claude, [#354](https://github.com/chardet/chardet/issues/354) <https://github.com/chardet/chardet/pull/354>_)

Accuracy:

• Accuracy improved from 98.6% to 99.3% (2499/2517 files) through a combination of training and scoring improvements:

  • Eliminated train/test data overlap by content-fingerprinting test suite articles and excluding them from training data ([#351](https://github.com/chardet/chardet/issues/351) <https://github.com/chardet/chardet/pull/351>_)
  • Added MADLAD-400 and Wikipedia as supplemental training sources to fill gaps left by exclusion filtering ([#351](https://github.com/chardet/chardet/issues/351) <https://github.com/chardet/chardet/pull/351>_)
  • Improved non-ASCII bigram scoring: high-byte bigrams are now preserved during training (instead of being crushed by global normalization), and weighted by per-bigram IDF so encoding-specific byte patterns contribute proportionally to how discriminative they are ([#352](https://github.com/chardet/chardet/issues/352) <https://github.com/chardet/chardet/pull/352>_)
  • Added encoding-aware substitution filtering: character substitutions during training now only apply for characters the target encoding cannot represent
  • Increased training samples from 15K to 25K per language/encoding pair (Dan Blanchard <https://github.com/dan-blanchard>_ via Claude)

... (truncated)

• d9ae78d docs: changelog for 7.4.1
• 2a54c68 Return utf-16/utf-32 (not -le/-be) when a BOM is present (#365)
• c63c632 Address GitHub code quality findings and add missing test coverage
• 1ad8e6a Revert "Add PyInstaller hook to collect mypyc shared runtime library (#359)" ...
• 7fb0563 Add PyInstaller hook to collect mypyc shared runtime library (#359)
• 2d75e6d Link to blogpost in README
• e37cf3c fix: prevent dirty-tree version in Windows mypyc wheel builds
• f9f5af2 Fix a couple errors in the changelog
• 53755de chore: add .superpowers/ to .gitignore
• 3a20df6 docs: update README examples with correct outputs
• Additional commits viewable in compare view