Skip to content
AI Model Radar

granite docling 258M

ibm-granite/granite-docling-258M
Open Source · chat · open-weights
Open Alert me on changes
Context
8K
Max output
—
Weights
Open
API $/1M
—
Modalities
text · image
Released
19 May 2025
License: apache-2.0 · ibm-granite/granite-docling-258M
Download image Share on X Share on LinkedIn
AI summary
● machine-written

IBM Granite Docling 258M: Open-source document AI model

Granite Docling 258M is a compact vision-language model designed for document conversion, parsing PDFs, slides, and scanned pages into structured formats. It integrates vision and language processing into a single 258M-parameter model with a Granite 165M backbone and SigLIP2 vision encoder. The model outputs structured DocTags to preserve layout, tables, code, and equations across document types.

What's new
  • Compact 258M-parameter design succeeds SmolDocling
  • Outputs structured DocTags for layout and content preservation
  • Supports text and image inputs across PDFs, slides, and scanned pages
  • Includes measurable accuracy gains for OCR, equations, and tables
Best for
Document conversion and extraction workflowsPDF to Markdown/HTML/structured output conversionPreserving layout fidelity and content integrityProcessing scanned documents and presentation slides
Sources

Source: https://huggingface.co/ibm-granite/granite-docling-258M