NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #2152 most downloaded on PyPI
SDK and CLI for parsing PDF, DOCX, HTML, and more, to a unified document representation for powering downstream workflows such as gen AI applications.
Latest stable
2.150.1
Last release 4 days ago
14 Nov 2026
Ships on a steady schedule
a new release about every 9 days
Nearly every release is documented
notes for 60 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
2 years old
241 releases · first in 2024
Drag across the ruler to compare, or tap a release to open it.
Upgrade to RapidOCR 3.x (#2088) (`3f60a0f`)
Vllm extra only for linux x86_64 (#2126) (`488f6cd`)
Your coding agent can read these notes before it upgrades. Set up the MCP server →
CLI: Option to download arbitrary HuggingFace model (#2123) (`cdf079d`)
New code formula model (#2042) (`d2494da`)
Add backend for METS with Google Books profile (#1989) (`31087f3`)
Add convert_string to document-converter (#2069) (`b09033c`)
HTML: Concatenation of child strings in table cells and list items (#1981) (`5132f06`)
Add option to control empty clusters in layout postprocessing (#1940) (`a436be7`)
cca05c4)e1e3053)95e7096)c5fb353)Layout model specification and multiple choices (#1910) (`2b8616d`)
Introduce LayoutOptions to control layout postprocessing behaviour (#1870) (`ec6cf6f`)
ec6cf6f)56a0e10)598c9c5)ae39a94)Leverage new list modeling, capture default markers (#1856) (`0533da1`)
Updated granite vision model version for picture description (#1852) (`d337825`)
Support audio input (#1763) (`1557e7c`)
1557e7c)861abcd)215b540)d26dac6)1350a8d)dd7f64f)dbab30e)Make Page.parsed_page the only source of truth for text cells, add OCR cells to it (#1745) (`7d3302c`)
7d3302c)df14022)f28d23c)b886e4d)7a275c7)6613b9e)e979750)f7f3113)9dbcb3d)a2b83fe)Remove typer and click constraints (#1707) (`8846f1a`)
Simplify dependencies, switch to uv (#1700) (`cdd4018`)
Add visualization of bbox on page with html export. (#1663) (`b356b33`)
ocr: Auto-detect rotated pages in Tesseract (#1167) (`45265bf`)
Add textbox content extraction in msword_backend (#1538) (`12a0e64`)
f4d9d41)0e00a26)f2e9c07)98b5eeb)Improve parallelization for remote services API calls (#1548) (`3a04f2a`)
AsciiDoc header identification (#1562) (#1563) (`4046d0b`)
Add smoldocling in download utils (#1577) (`127e386`)
127e386)776e7ec)f1658ed)7c70573)4ab7e9d)cc45396)976e92e)94d66a0)Add tutorial using Milvus and Docling for RAG pipeline (#1449) (`a2fbbba`)
ed20124)8012a3e)fa7fc9e)c2470ed)64918a8)995b3b0)88948b0)a7dd59c)293c28c)01fbfd5)a026b4e)cli: Add option for html with split-page mode (#1355) (`c0ba88e`)
c0ba88e)eef2bde)c605edd)7e40ad3)0de70e7)415b877)2503999)6b696b5)Handle tags as code blocks (#1320) (`0499cd1`)
<code> tags as code blocks (#1320) (0499cd1)bfcab3d)14e9c0c)dc3bf9c)d2d6874)b3d111a)Fixes tables when using OCR (#1261) (`7afad7e`)
Word-level pdf cells for tables (#1238) (`8bd71e8`)
Improve HTML layer detection, various MD fixes (#1241) (`9210812`)
converter: Cache same pipeline class with different options (#1152) (`825b226`)
SmolDocling: Support MLX acceleration in VLM pipeline (#1199) (`1c26769`)
1c26769)b454aa1)2f72167)f5adfb9)0b707d0)Add factory for ocr engines via plugins (#1010) (`6eaae3c`)
6eaae3c)3960b19)772487f)6eb718f)f94da44)0945973)aa92a57)Use new TableFormer model weights and default to accurate model version (#1100) (`eb97357`)
Proper handling of orphan IDs in layout postprocessing (#1118) (`c56ab3a`)
Enable locks for threadsafe pdfium (#1052) (`8dc0562`)
[Experimental] Introduce VLM pipeline using HF AutoModelForVision2Seq, featuring SmolDocling model (#1054) (`3c9fe76`)
3c9fe76)ab683e4)e197225)1b0ead6)Implement new reading-order model (#916) (`c93e369`)
Runtime error when Pandas Series is not always of string type (#1024) (`6796f0a`)
Support cuda:n GPU device allocation (#694) (`77eb77b`)
Add support for CSV input with new backend to transform CSV files to DoclingDocument (#945) (`00d9405`)
00d9405)2716c7d)5101e25)af19c03)c47ae70)Add content_layer property to items to address body, furniture and other roles (#735) (`cf78d5b`)
Describe pictures using vision models (#259) (`4cc6e3e`)
New artifacts path and CLI utility (#876) (`ed74fe2`)
90b766e)9114ada)722a6eb)5ad6de0)Expose equation exports (#869) (`6a76b49`)
6a76b49)70d68b6)d727b04)4df085a)5ac2887)94751a7)0cd81a8)eff16b6)b1cf796)2c037ae)2a1f8af)bccb022)fea0a99)CLI: Expose code and formula models in the CLI (#820) (`6882e6c`)
6882e6c)95b293a)rec_keys_path in RapidOcrOptions to support custom dictionaries (#786) (5332755)3be2fb5)5aed9f8)adf6353)a112d7a)6875913)5139b48)4d41db3)8a4ec77)b885b2f)c2ae1cc)New document picture classifier (#805) (`16a218d`)
16a218d)88a0e66)3213b24)8543c22)a458e29)670a08b)Improve OCR results, stricten criteria before dropping bitmap areas (#719) (`5a060f2`)
Added http header support for document converter and cli (#642) (`0ee849e`)
Create a backend to transform PubMed XML files to DoclingDocument (#557) (`fd03480`)
Updated Layout processing with forms and key-value areas (#530) (`60dc852`)
Introduce support for GPU Accelerators (#593) (`19fad92`)
Add timeout limit to document parsing job. DS4SD#270 (#552) (`3da166e`)
Docling-parse v2 as default PDF backend (#549) (`aca57f0`)
Expose new hybrid chunker, update docs (#384) (`c8ecdd9`)
c8ecdd9)3e073df)eb7ffcd)py.typed marker file (#531) (9102fe1)0d11e30)b730b2d)c830b92)8ada0bc)Improve handling of disallowed formats (#429) (`34c7c79`)
ParserError EOF inside string (#470) (#472) (`c90c41c`)
c90c41c)d3f84b2)5ba3807)33cff98)d487210)8ccb3c6)One column per month.