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 5 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Nothing withdrawn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.
service: Add S3Coordinates.max_num_elements (#3565) (`d3159d7`)
d3159d7)057241c)271d3c7)8c61a78)Your coding agent can read these notes before it upgrades. Set up the MCP server →
service: Updates for service client, batch endpoint support, and cleanup (#3519) (`b613414`)
37c6a1e)b7b01f1)efcc193)854b2b1)0a22016)d2b4e74)953b298)f971048)3d456f8)Improve missing FFmpeg error for ASR (#3478) (`11a1bb5`)
Add threaded docling-parse (v6) PDF backend (#3377) (`3c26f5a`)
Add scoped settings context manager (#3440) (`d907d21`)
security: Add GitHub Private Vulnerability Reporting (#3416) (`24af7f6`)
eceedc2)5fadc6d)b5f2e53)336f942)81afde8)52d4887)ab6aa05)038b991)bcd5509)694cf0c)0c31706)64ddeb6)aba7f15)eb6e1e6)2bb0fa6)6b3322e)vlm: Upgrade Granite Vision model to 4.1 for table + chart extraction (#3382) (`24f2d14`)
Extend the kserve-triton OCR model to have multi-lingual support (#3368) (`8b67fae`)
docx: Extract VML images with v:imagedata elements (#3343) (`2ddaa3b`)
c1dbac2)5e161ac)c190ba2)cd0cb69)9813190)3a3c8f6)65ef180)075fa69)Implement GraniteVisionTableStructureModel for VLM-based table extraction (#3323) (`1569e42`)
101233e)c761512)3bab6b4)8274892)043ed2d)Explicit TikZ environment handling in LaTeX backend (#3187) (`a15c16e`)
service: Establish client SDK for docling serve (#3264) (`42157a3`)
vlm: Add Nanonets OCR2 onboarding (#3274) (`9970d1e`)
d431224)27d3cf4)a6aeddf)6cb1bc0)e4fd937)9c3ab93)ab5254d)Support for GraniteVision v4 (#3217) (`fd83420`)
Upgrade to transformers v5 (#3200) (`d2c6357`)
Implementation of HTML backend with headless browser (#2969) (`1c74a9b`)
e36125b)a0fc3c9)90d6dd4)fdf5e20)f0e3d1d)5473e07)9abf0fd)Route plain-text and Quarto/R Markdown files to the Markdown backend (#3161) (`96d7c7e`)
2f7c09e)afb4bb6)8ae0974)a39317a)412af62)53a5f80)1e841eb)Add the VllmCudaGraphMode (#3125) (`f950679`)
Add fact metadata and linkbase relationships for XBRL (#3084) (`7952efe`)
Add support for TableFormer v2 (#3013) (`4ccd1d4`)
4ccd1d4)3d90778)80f75b8)5188180)1192714)ee16285)Track vlm_inference time for mlx_model pipeline (#3060) (`38c4bb2`)
38c4bb2)cfc6636)cd9dd10)56f06fe)f7cb304)56eb127)68336c2)Export to WebVTT format (#3036) (`d276e60`)
859c302)6198e69)477359b)90ce93d)a3d2b4b)Create a backend parser for XBRL instance reports (#3017) (`334ba6e`)
Security vulnerabilities with XML External Entity and related attacks (#3009) (`576bada`)
Inference engines abstraction for object detection model family with HF Transformers and ONNX runtime (#2959) (`14e474c`)
14e474c)e6ccb8b)d4c8713)9721321)ae4fdbb)Add chart extraction models (#2848) (`fe45c71`)
Webvtt and source tracker (#2787) (`0602a7c`)
Drop support for Python 3.9 (#2905) (`7f38658`)
Off-by-one error for page indexing in vlm_pipeline (#2902) (`08f49e2`)
New picture classifier v2.0 (#2889) (`43badc3`)
43badc3)ac16a26)00273f6)1b4d82d)2fe9def)daf2bc6)Support for DeepSeek-OCR in VLM pipeline (#2798) (`19af03f`)
Enrichment annotations in the new meta format (#2859) (`aab3ff5`)
aab3ff5)2b83fdd)cbc6537)d9295df)a0530a2)3ef4525)ffafe58)ed57089)Add preset for using granite-docling via vllm and other apis (#2792) (`241d19e`)
Add YAML output format to CLI (#2768) (`da7678a`)
Clear word/char cells when force_full_page_ocr is used (#2738) (`1df0560`)
1df0560)edbabfc)609069d)d007ba0)aebe25c)c97715f)experimental: Add experimental TableCropsLayoutModel (#2669) (`1344362`)
1344362)ad97e52)6ef4ffd)54cd6d7)e580554)examples: Remove deprecation warnings with export_to_dataframe (#2638) (`f552862`)
2087c6b)6fb9a5f)463a3fd)da4c2e9)ce5a099)b216ad8)03e7c7d)8af228f)d549445)ac9fc58)f552862)Add the Image backend (#2627) (`3495b73`)
Default to EasyOCR in Python 3.14 (#2605) (`5c27567`)
vlm: Track generated tokens and stop reasons for VLM models (#2543) (`6a04e27`)
Extract response from api_image_request in picture description (#2571) (`8360aa5`)
Use threading in the standard pipeline and move old behavior to legacy (#2452) (`268d027`)
vlm: Add num_tokens as attribtue for VlmPrediction (#2489) (`b6c892b`)
pdf: Support for password-protected PDF documents (#2499) (`bbe82a6`)
bbe82a6)a30e6a7)657ce8b)4227fcc)b66624b)docx: Process drawingml objects in docx (#2453) (`1682993`)
Avoid downloading easyocr models by default (#2454) (`688a7df`)
AutoOCR model selecting the best OCR model available and deprecating the usage of EasyOCR (#2391) (`f7244a4`)
f7244a4)f11f8c0)db985bb)cce18b2)ee55013)b5f7fef)0610d01)9705f40)markdown: Setext heading support (#2359) (`ee73ffa`)
Repetition-based StoppingCriteria for GraniteDocling (#2323) (`1e9dc43`)
1e9dc43)c803abe)68ae7cc)654c70f)9d67bb9)Rich tables for MSWord backend (#2291) (`e2482a2`)
Add granite-docling model (#2272) (`17afb66`)
Address deprecation warnings of dependencies (#2237) (`c696549`)
Updating default parameters to get better performance with docling-parse (#2208) (`b49d1ad`)
Heron layout model as new default (#1971) (`e38aa0f`)
[Beta] Extraction with schema (#2138) (`9f4bc5b`)
9f4bc5b)a283ccf)4d94e38)9f0286b)9904d14)One column per month.