PaddleOCR-VL vs Kraken
Pick Kraken for historical manuscripts, early prints and right-to-left scripts, where you can train layout and recognition on your own material. Pick PaddleOCR-VL for modern multilingual pages with tables, formulas and charts that you want as Markdown or JSON without training. The main split is that Kraken is a trainable toolkit for historical material, while PaddleOCR-VL is a ready-to-run page parser.
PaddleOCR-VLpaddle/paddleocr-vl | Krakenkraken/kraken | |
|---|---|---|
| Type | Open source | Open source |
| Price | 6 credits / page | 1 credit / page |
| Per 1,000 pages | $6.00 | $1.00 |
| Typical speed | 1.3 s / page | 1.5 s / page |
| Best for | CJK documents | Historical documents and manuscripts |
| Inputs | PDF, images | images |
| Languages | 109 languages | Many scripts, incl. right-to-left |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | ✓ Yes | No |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| 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 PaddleOCR-VL if
- It parses text, tables, formulas and charts from a full page into Markdown and JSON without any training.
- It covers 109 languages, including Cyrillic, Arabic, Devanagari and Thai scripts, in a single model.
- At about 0.9B parameters it runs on much smaller GPUs than 7B document VLMs like olmOCR, and it has published scores (96.33 on OmniDocBench v1.6, 80.0 on olmOCR-Bench).
Choose Kraken if
- It is built for historical manuscripts and early prints, with right-to-left, BiDi and top-to-bottom support.
- Layout analysis, reading order and character recognition are all trainable on your own material.
- It writes ALTO, PageXML, abbyyXML and hOCR, which digital humanities tools use, and community-trained models are available on Zenodo.
Key differences
| PaddleOCR-VL | Kraken | |
|---|---|---|
| Main target | Multilingual scans, academic papers, tables and charts in reports | Historical manuscripts, early prints, Arabic, Hebrew and other RTL scripts |
| Setup effort | Run a pretrained model via PaddleOCR, vLLM or transformers | Usually needs training on your own material for good accuracy |
| Output formats | Markdown, JSON | ALTO, PageXML, hOCR, plain text, bounding boxes |
| Platform and hardware | NVIDIA GPU recommended; CPU possible but slow; no minimum VRAM stated | Linux or macOS, x64 or ARM, Python 3.10 to 3.13; no Windows support |
| Licence and price | Apache-2.0, free, you pay for compute | Apache-2.0, free, you pay for compute |
| Published benchmarks | OmniDocBench v1.6: 96.33 (authors); olmOCR-Bench: 80.0 (independent) | Not published |
| Known failure | Version 1.6 regresses Russian and Kazakh OCR on born-digital PDFs, hangs on blurry images, and decodes about 11x slower because use_cache is off | Install can fail with a dependency conflict, and learned line order is lost in segmentation and text output |
Sources
- Hugging Face model card
- PaddleOCR-VL-1.6 model card
- olmOCR-Bench table
- Kraken GitHub README
- Kraken docs
- Issue #820
What each one is
PaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-VL detailsKraken
OCR engine for historical documents and manuscripts, with trainable models for unusual scripts.
Kraken 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": "paddle/paddleocr-vl","fallbacks": ["kraken/kraken"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR-VL alternativesKraken alternativesBest OCR models in 2026