Kraken vs Granite Docling
Pick Kraken for historical manuscripts, early prints and right-to-left scripts such as Arabic and Hebrew, where you can train layout and recognition on your own material. Pick Granite Docling for English PDFs with tables, code and equations, especially inside a Docling pipeline. The main split is material: Kraken targets old and non-Latin documents and needs training, while Granite Docling targets modern English documents and runs from a 258M model.
Krakenkraken/kraken | Granite Doclingibm/granite-docling | |
|---|---|---|
| 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 | 1.5 s / page |
| Best for | Historical documents and manuscripts | Small-model document conversion |
| Inputs | images | PDF, images |
| Languages | Many scripts, incl. right-to-left | English, experimental CJK and Arabic |
| 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 is what Kraken is built for.
- You need right-to-left, BiDi or top-to-bottom scripts such as Arabic and Hebrew.
- You want to train layout analysis, reading order and character recognition yourself, and can reuse community models from the public Zenodo repository.
Choose Granite Docling if
- Your documents are English PDFs with tables, code or equations, where the model card reports table structure TEDS of 0.97 on FinTabNet and equation F1 of 0.968.
- You already use Docling or want a small 258M model that converts on CPU or a small GPU.
- You want Markdown, HTML or LaTeX output instead of ALTO or PageXML.
Key differences
| Kraken | Granite Docling | |
|---|---|---|
| Target material | Historical documents and non-Latin scripts | English PDFs with tables, code and equations |
| Licence and price | Apache-2.0, free, self-hosted (you pay for compute) | Apache-2.0, free, self-hosted (you pay for compute) |
| How you run it | Python library installed with pip on Linux or macOS, Python 3.10 to 3.13 | Self-hosted VLM, typically inside the Docling library |
| Output formats | ALTO, PageXML, abbyyXML, hOCR, plain text, bounding boxes | DocTags, Markdown, HTML, LaTeX, bounding boxes |
| Benchmarks | Not published | Authors report table TEDS 0.97 (FinTabNet), full-page OCR F1 0.84, OCRBench 500 |
| Languages | Right-to-left, BiDi and top-to-bottom scripts supported | English primary; Japanese, Arabic and Chinese marked experimental |
| Known failure | No Windows support; install dependency conflict in issue #816; learned line order lost in output (issue #820); usually needs training for good accuracy | Repetitive loops on page headings and invoices; may output only exclamation marks on GPUs without bfloat16 such as T4 unless run in float32 |
Sources
- Kraken GitHub README
- Kraken docs
- Zenodo OCR models
- Issue #820
- Granite-Docling model card
- HF discussion #24 repetition
What each one is
Kraken
OCR engine for historical documents and manuscripts, with trainable models for unusual scripts.
Kraken detailsGranite Docling
Compact IBM vision model that converts pages to DocTags for the Docling pipeline.
Granite Docling 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": ["ibm/granite-docling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Kraken alternativesGranite Docling alternativesBest OCR models in 2026