GPT vs Kraken
Pick GPT for handwritten notes, custom JSON extraction and document question answering through a hosted API. Pick Kraken for historical manuscripts, early prints and right-to-left or other non-Latin scripts that you can train on and run yourself. The main split is a paid, hosted, general model versus a free, trainable, self-hosted OCR system for historical material.
GPTopenai/gpt | Krakenkraken/kraken | |
|---|---|---|
| Type | Vision model | Open source |
| Price | 11 credits / page | 1 credit / page |
| Per 1,000 pages | $11.00 | $1.00 |
| Typical speed | 2.4 s / page | 1.5 s / page |
| Best for | Messy scans, custom instructions | Historical documents and manuscripts |
| Inputs | PDF, images | images |
| Languages | 50+ languages | Many scripts, incl. right-to-left |
| Max pages per request | 100 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | ✓ Yes |
| Math | ✓ Yes | No |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | 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 GPT if
- It is prompted to transcribe or extract data, so you can pull fields into custom JSON with reasoning.
- An HN user reported about 5% errors on messy cursive, versus about 30% for Mistral OCR.
- It runs as a hosted API with files up to 50 MB per request, so there is nothing to install or train.
Choose Kraken if
- It is built for historical documents and non-Latin scripts, with right-to-left, BiDi and top-to-bottom support.
- Layout analysis, reading order and character recognition are fully trainable on your own material.
- It is Apache-2.0 and free to self-host, and it returns bounding boxes plus ALTO, PageXML and hOCR.
Key differences
| GPT | Kraken | |
|---|---|---|
| Licence | Proprietary | Apache-2.0 |
| How you run it | Hosted API from OpenAI | Python library on Linux or macOS (x64 or ARM, Python 3.10 to 3.13). No Windows support. |
| Price | Token-based per model. Page images add input tokens, so cost per page is not fixed. | Free. You pay for your own compute. |
| Output formats | Markdown, JSON, plain text. No bounding boxes or confidence scores in the standard response. | ALTO, PageXML, hOCR, plain text, bounding boxes |
| Benchmark | GPT-5.2 scored 85.50 on OmniDocBench v1.5 and 86.59 on v1.6. Gemini 3 Pro scored 90.33 on v1.5 in the same paper. | Not published |
| Best document types | Handwritten notes, document question answering | Historical manuscripts, early prints, Arabic, Hebrew and other RTL scripts |
| Known problems | Not suited to cheapest bulk OCR or pipelines that need word coordinates. | Install can fail with a dependency conflict (issue #816). Learned line order is lost in segmentation and text output (issue #820). Usually needs training on your own material for good accuracy. |
Sources
- OpenAI file inputs docs
- PaddleOCR-VL-1.5 paper (GPT-5.2 eval)
- HN Mistral OCR thread
- Kraken GitHub README
- Issue #816
- Issue #820
What each one is
Kraken
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.
PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?
Use one, fall back to the other
{"model": "openai/gpt","fallbacks": ["kraken/kraken"],"file_url": "https://example.com/document.pdf"}