REDDIT · REDDIT// 21d agoOPENSOURCE RELEASE

Kreuzberg 4.5.0 integrates Docling layout model

Kreuzberg v4.5.0 brings native document structure and table understanding by embedding Docling's RT-DETR v2 model into its Rust-native pipeline. The integration delivers comparable accuracy to Docling but runs nearly three times faster with significantly lower memory overhead and no Python dependency.

// ANALYSIS

By porting Docling's excellent layout models to a Rust-native pipeline, Kreuzberg offers a compelling, high-performance alternative for heavy document processing pipelines.

–Replacing Python with Rust and using pdfium for character-level text extraction yields a massive 2.8x performance boost and lowers memory requirements
–Cross-language support (bindings for 12 languages including Python, JS, Go, Java) means you aren't forced into a Python stack just for document AI
–SLANet-Plus integration specifically targets table structure prediction, which is notoriously difficult in PDF extraction
–The project proves that leveraging permissive open-source models (Apache) inside highly optimized execution environments is a winning strategy for infrastructure tools

// TAGS

kreuzbergdoclingdata-toolsopen-source

DISCOVERED

21d ago

2026-03-22

PUBLISHED

21d ago

2026-03-22

RELEVANCE

8/ 10

AUTHOR

Eastern-Surround7763