Models & Pipelines

PROOF offers two processing models for different input types, accuracy requirements, and budgets.

Basic model (ocr_model=basic)

The basic model uses Docling for PDFs with a text layer, and Tesseract OCR for scanned documents and images. It is cost-effective and fast.

• Best for: Clean digital PDFs with embedded text
• Speed: Fast — text extraction without LLM calls
• Cost: $0.001/page (1 milli-USD), 100 free pages per day
• Limitations: No audio/video support, lower accuracy for complex layouts

curl -X POST https://your-domain.com/api/v1/jobs/upload \
  -H "X-API-Key: chu_live_xxxxxxxx" \
  -F "files=@document.pdf" \
  -F "ocr_model=basic"

Premium model (ocr_model=premium)

The premium model uses high performance AI vision model for all pages, providing superior accuracy for complex documents, scanned images, handwritten text, rotated pages, and tables.

• Best for: Scanned documents, handwritten text, complex tables, images, audio/video
• Speed: Moderate — one LLM call per page
• Cost: $0.005/page (5 milli-USD), no free quota
• Features: Audio ($0.001/sec), video ($0.005/sec), Docling refinement option

curl -X POST https://your-domain.com/api/v1/jobs/upload \
  -H "X-API-Key: chu_live_xxxxxxxx" \
  -F "files=@document.pdf" \
  -F "ocr_model=premium"

OCR engine selection

For the premium model, you can specify the OCR engine used for text extraction:

Engine	Flag	Notes
easyocr	Default	Good balance of speed and accuracy
tesseract	ocr_engine=tesseract	Fast, widely supported
rapidocr	ocr_engine=rapidocr	Optimized for CJK text

curl -X POST https://your-domain.com/api/v1/jobs/upload \
  -H "X-API-Key: chu_live_xxxxxxxx" \
  -F "files=@document.pdf" \
  -F "ocr_model=premium" \
  -F "ocr_engine=rapidocr"

Choosing a model

Scenario	Recommended model
Clean digital PDF with text layer	basic
Scanned document	premium
Handwritten text	premium
Image with table	premium
Audio/video	premium (required)
Budget-conscious, high volume	basic (100 free pages/day)