New features, improvements, and breaking changes for every PDFluent release.
v1.1.0
Current2026-10-17
featureStructured extraction for every channel: Markdown, blocks, words with boxes, and chunks for retrieval pipelines. Preview.
featureOptional balanced extraction quality runs a layout model and a table-structure model locally through ONNX Runtime 1.23.2 [B49, 2026-10-10]. Native only.
featureModels install explicitly with pdfluent models install layout tables; nothing downloads at run time.
featureImage extraction on the plain document API, a compat layer that follows part of the PyMuPDF API (coverage is partial and behaviour can differ in places; it is an independent implementation, not PyMuPDF), and a full command-line interface.
improvementWords are joined by default; ligatures come out as letters; /ActualText replacement is opt-in.
improvementBlock is #[non_exhaustive] and gained confidence and grid.
improvementpdf-syntax and pdf-interpret move to 0.6.0.
improvementJava: 1.1.0 is served from the PDFluent Maven repository at https://maven.pdfluent.com. Maven Central carries 1.0.0; 1.1.x follows there later.
featurePDFluent SDK 1.0.0 is generally available as of 2026-09-16.
featureEvery binding channel moves to 1.0.0: the Rust crate pdfluent, PyPI pdfluent, npm @pdfluent/node and @pdfluent/sdk-wasm, Maven com.pdfluent:pdfluent, and NuGet PDFluent.
featureThe npm @pdfluent/node package ships with per-platform packages alongside the main package.
improvementA backslash in a page's text no longer causes the PDF/A pass to drop the rest of that page's text.
improvementSpaces no longer render as ")" on subset fonts.
improvementLow character codes survive where the font's /Differences names them, for example TeX subset fonts.
improvementsign_pdf_incremental now works; it used to fail on every input with "PDF has no pages". The previous revision survives byte for byte.
improvementA certified signature now reports its own DocMDP level; the reader had looked up the wrong key. Form-field locks (/FieldMDP) are seen again.
improvementThe PNG Average predictor encoder output can now be decoded by its own decoder.
improvementSeveral of these fixes were found by measuring text retention against the source document as a second axis next to PDF/A validation.
v0.9.0
April 2026
featurePDF/A-1b, PDF/A-2b, PDF/A-3b validation and conversion
featurePDF/UA-1 accessibility compliance checking
featureXFA form support — parse, FormCalc evaluation, flatten to standard PDF
featureWASM build target — compile to WebAssembly, run in the browser
featureAES-256 encryption with owner/user password and permission flags
featureDigital signature creation and validation
featureRenderer tested against a real-world PDF corpus
improvementSSIM visual accuracy improvements across all page types