Ai

Docling v2.92.0 Expands Multilingual OCR and Document Parsing

Docling v2.92.0 Expands Multilingual OCR and Document Parsing

Docling v2.92.0 delivers a focused update centered on broader document understanding and more flexible deployment. The release expands OCR with multilingual support in the kserve-triton model, improves DOCX parsing with checkbox handling, introduces a new modular docling-slim package, and adds ResponseFormat.DOCLANG support in the VLM pipeline. It also tightens conversion reliability with fixes for malformed PPTX picture shapes, DOCX OMML conversion edge cases, and configurable VLLM model implementation settings.

What Changed

One of the most notable additions in v2.92.0 is multilingual support for the kserve-triton OCR model. This extends Docling’s OCR capabilities for teams processing document sets across multiple languages, making the platform more practical for international and enterprise workflows.

The DOCX pipeline also gains checkbox parsing support, which improves structured extraction from forms, checklists, and other business documents where checkbox state carries meaning. In parallel, the release introduces the modular docling-slim package, giving users a lighter deployment option that may better suit resource-constrained environments or modular packaging strategies.

Another key feature is the addition of ResponseFormat.DOCLANG and a corresponding parsing branch in the VLM pipeline. This suggests continued investment in more specialized output formatting and multimodal document-processing flows, which could matter for downstream automation and AI-driven extraction pipelines.

On the reliability side, Docling now skips malformed picture shapes in PPTX files instead of aborting conversion entirely. DOCX handling is also improved through a fix for OMML conversion failures tied to unsupported limit functions. The release further makes the VLLM model_impl configurable, adding flexibility for deployment and inference environments.

Why It Matters

This version matters because it strengthens Docling’s usefulness in real-world document automation environments where input quality, language diversity, and file complexity are often inconsistent. Multilingual OCR and richer DOCX parsing directly improve extraction fidelity for enterprise content pipelines, while the new slim package can help teams optimize packaging and runtime overhead.

The stability fixes are equally important. By preventing single malformed elements from breaking entire conversions, v2.92.0 reduces operational friction and improves throughput for organizations processing large volumes of presentations and office documents. Combined with added VLM and VLLM flexibility, the update positions Docling as a more resilient and adaptable platform for AI-driven document workflows.

Official Source: https://github.com/docling-project/docling/releases/tag/v2.92.0