Kraken vs SmolDocling
Pick Kraken for historical manuscripts, early prints and non-Latin or right-to-left scripts, where you can train layout analysis and recognition on your own material. Pick SmolDocling only for experiments with a tiny document VLM on English pages, since it is a preview release and its authors point users to Granite-Docling. The main split is the job: Kraken is a trainable toolkit for historical material, SmolDocling is a 256M model that converts modern-style pages to structured output.
Krakenkraken/kraken | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 1 credit / page |
| Per 1,000 pages | $1.00 | $1.00 |
| Typical speed | 1.5 s / page | 1.0 s / page |
| Best for | Historical documents and manuscripts | Very cheap page conversion |
| Inputs | images | PDF, images |
| Languages | Many scripts, incl. right-to-left | Mainly English |
| Max pages per request | 500 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose Kraken if
- Your material is historical manuscripts or early prints, which Kraken is built for.
- You need Arabic, Hebrew or other right-to-left scripts, with BiDi and top-to-bottom support.
- You want ALTO, PageXML or hOCR output for digital humanities tools, and you can train layout, reading order and recognition on your own data.
Choose SmolDocling if
- You want one 256M-parameter model that handles text, tables (OTSL), code, formulas and charts with location tags.
- You need DocTags, Markdown, HTML, JSON or LaTeX output from English pages on modest hardware.
- Speed matters in an experiment: about 0.35 seconds per page on an A100 with vLLM.
Key differences
| Kraken | SmolDocling | |
|---|---|---|
| What it is | Open-source OCR system with trainable layout analysis and recognition | 256M-parameter document conversion VLM from Hugging Face and IBM that outputs DocTags for Docling |
| Licence | Apache-2.0 | Apache-2.0 per the model summary; the metadata tag says CDLA-Permissive-2.0 |
| How you run it | Python library via pip on Linux or macOS (x64 or ARM, Python 3.10 to 3.13); no Windows support | Self-hosted model; you pay for compute |
| Output formats | ALTO, PageXML, hOCR, plain text, bounding boxes | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Languages | Historical and non-Latin scripts, including right-to-left, BiDi and top-to-bottom | English only |
| Benchmarks | Not published | Table TEDS structure 0.82 on FinTabNet 150dpi, full-page OCR F1 0.80, OCRBench 338 (author-reported) |
| Known failure | Install can fail with a dependency conflict (issue #816); learned line order is lost in segmentation and text output (issue #820) | Endless token repetition and missing closing tags; output loops on some images, making generation very slow (Docling issue #1515) |
Sources
- Kraken GitHub README
- Kraken docs
- Kraken issue #820
- SmolDocling model card
- SmolDocling paper
- Docling issue #1515
What each one is
Kraken
OCR engine for historical documents and manuscripts, with trainable models for unusual scripts.
Kraken detailsSmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "kraken/kraken","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Kraken alternativesSmolDocling alternativesBest OCR models in 2026