What's New

Docling for IBM watsonx is updated continuously. Updates roll out automatically, so you get new features and quality improvements without changing anything on your side. The latest update is listed first.

October 2026

DocLang by default in the Workbench. New tasks now include DocLang (.dclx) output, and a DocLang viewer action opens the result in the DocLang Viewer in one click. See Explore with DocLang.
Apple Keynote and Numbers support. .key presentations and .numbers spreadsheets can now be converted. Their charts are recognized as charts with their data, and pictures placed on Numbers sheets are kept.
Better reading order for PDFs. A reworked reading-order algorithm gives more natural text flow on complex multi-column layouts.
Better document structure. Tables in Markdown files keep bold, italic and links inside their cells, and heading levels are now also inferred from "Division" and "Subdivision" markers common in legal and regulatory texts.
Faster chunking for large documents. Splitting documents with many elements into chunks takes much less time.
Clearer task tracking in the Workbench. Task status is updated as soon as a task completes. Document-level errors and skipped documents are now shown directly in the task view.
Drag and drop in the Workbench. You can now drop a file straight into the new-task dialog.

September 2026

New connectors: Box and Databricks Volumes. You can now read documents from these platforms and write results back to them with the batch endpoint.
More input formats: MHTML web archives and AFP, plus complete Apple Pages (.pages) support.
Faster PDF processing. PDF parsing and page rendering are up to 3.8× faster, and table matching and layout analysis are faster too.
Headings from PDF outlines. Heading levels now also use the PDF's bookmarks and outline, which makes the document hierarchy more reliable.
Redesigned conversion options in the Workbench. Options are grouped by processing stage and each one is described. Output formats are easier to pick, and more file types can be uploaded.

August 2026

Kafka streaming target. Stream document chunks straight into Apache Kafka or Confluent topics, ready for event-driven RAG pipelines. You can also configure it from the Workbench.
New connectors: SharePoint and OneDrive, and AstraDB. Read documents from Microsoft 365, or write chunks straight into an AstraDB vector collection.
Batch tasks in the Workbench. Create batch tasks with several sources and targets from the Workbench, including the new connectors.
Better heading hierarchy. Heading levels in PDFs are now inferred from font weight, style and case. A new option turns this inference on or off through the API.
More input formats: Outlook .msg emails, Apple Pages and EBCDIC text.
New docling-client Python package and TypeScript examples. docling-client is a lightweight client dedicated to the service. The Workbench now also shows TypeScript API examples.
Higher-fidelity page images. Page rendering is much more accurate, especially for colors, transparency, clipping and CJK text.

July 2026

Chunking and vector database targets. The batch endpoint can chunk documents (chunking_options) and write the chunks straight to a vector database such as OpenSearch. A single batch request can also write to several targets.
New connectors: Google Cloud Storage, Azure Blob Storage and Google Drive. The FileNet connector now also accepts specific document IDs.
DocLang output. Request "doclang" in to_formats to get your document as a DocLang .dclx archive. DocLang is the open, structured document format, and you can inspect it in the DocLang Viewer.
More input formats: EPUB e-books, BoxNote and DocLang archives.
Native Office charts. Charts in Word, Excel and PowerPoint files are now recognized as charts and their underlying data is kept.
Better structure from PDFs.
- Heading levels are inferred instead of being flattened.
- Headings also come from the PDF's table of contents.
- The language of code blocks is detected.
Confidence scores. Conversion results now report a confidence score for each document.
Faster conversions. A new generation of the PDF parser processes more pages per second, and OCR is faster too.
Refreshed Workbench. New landing, Tasks and Usage pages, and document previews load faster.

June 2026

General availability. Docling for IBM watsonx launched as a managed service.
- Single-document conversion from a file upload or a URL.
- The batch endpoint for large collections in Amazon S3, with results delivered to your own bucket or as temporary download links.
- The Workbench, with document preview, API key management, and ready-to-use Python and Java integration examples.
- Self-service trial and subscription plans.