Loading...
Test Corpus
11 test documents across 6 categories. Each engine is evaluated against these images with human-verified ground truth transcripts.
10 of 11 documents are synthetic
These were generated to represent common document types. They work for benchmarking accuracy but may not reflect your specific use case. We plan to replace them with real public-domain documents — see ocr-bench to contribute real test documents.
Receipt (2)Handwriting (2)Invoice (2)Table (2)Dense Text (2)Web UI Screenshot (1)
ReceiptSynthetic
Grocery Receipt
Synthetic receipt — tests retail-style dense numeric rows.
▶Ground truth transcript
CORNER STORE 123 Main Street Date: 2026-02-28 14:32 -------------------------------- Milk 2% 1x $3.49 Bread Wheat 1x $2.99 Eggs Large 12ct 1x $4.29 Coffee Dark Roast 1x $8.99 Bananas 1.2lb 1x $0.89 -------------------------------- Subtotal: $20.65 Tax (8.25%): $1.70 TOTAL: $22.35 -------------------------------- VISA ****4821 Thank you for shopping!
ReceiptSynthetic
Café Receipt
Synthetic receipt — tests retail-style dense numeric rows.
▶Ground truth transcript
CAFE BISTRO 456 Oak Avenue, Suite 2 02/28/2026 10:15 AM ================================ Cappuccino 1 $4.50 Croissant 2 $7.00 Orange Juice 1 $3.25 Avocado Toast 1 $9.50 ================================ Subtotal $24.25 Tip (18%): $4.37 TOTAL: $28.62 Paid: Apple Pay Have a great day!
HandwritingSynthetic
Handwritten Letter
Synthetic handwriting — tests cursive recognition.
▶Ground truth transcript
Dear Professor, I wanted to follow up on our discussion about the research proposal. The deadline is March 15 and I have completed the first three sections. Best regards, Alex
HandwritingSynthetic
Handwritten Notes
Synthetic handwriting — tests cursive recognition.
▶Ground truth transcript
Meeting Notes - Feb 28 - Review Q4 budget by Friday - Call vendor re: contract renewal - Update project timeline - Send report to Sarah - Book conference room for Wed - Prepare slides for demo Action: Sarah to confirm deadline with client
InvoiceSynthetic
Services Invoice
Synthetic invoice — tests tabular billing layouts.
▶Ground truth transcript
INVOICE #INV-2026-0472
From: Acme Solutions LLC
To: Widget Corp International
Date: February 28, 2026
Due Date: March 30, 2026
Description Qty Rate Amount
---------------------------------------------------
Web Development 40h $150.00 $6,000.00
UI/UX Design 20h $125.00 $2,500.00
Server Infrastructure 1 $500.00 $500.00
---------------------------------------------------
Subtotal: $9,000.00
Tax (10%): $900.00
TOTAL: $9,900.00
Payment Terms: Net 30
Bank: First National Bank Acct: 1234567890
InvoiceSynthetic
Products Invoice
Synthetic invoice — tests tabular billing layouts.
▶Ground truth transcript
INVOICE
Invoice No: 10582
Date: 2026-02-15
Bill To:
Johnson & Partners LLP
789 Corporate Blvd
Austin, TX 78701
Item Qty Price Total
----------------------------------------------
Office Chairs (Mesh) 5 $289.00 $1,445.00
Standing Desks 3 $549.00 $1,647.00
Monitor Arms 8 $79.00 $632.00
Cable Management Kit 8 $24.99 $199.92
----------------------------------------------
Subtotal: $3,923.92
Shipping: $150.00
TOTAL: $4,073.92
TableSynthetic
Benchmark Table
Synthetic table — tests column and row structure parsing.
▶Ground truth transcript
Engine Performance Benchmark Results Engine Latency(ms) Accuracy Cost/Page ---------- ----------- -------- --------- Tesseract 178ms 92.1% $0.0005 EasyOCR 86ms 88.7% $0.0005 docTR 908ms 94.3% $0.0005 GPT-4o 1600ms 98.9% $0.0100 Claude Sonnet 1738ms 99.1% $0.0080 Gemini Flash 798ms 97.2% $0.0030
TableSynthetic
Revenue Table
Synthetic table — tests column and row structure parsing.
▶Ground truth transcript
Monthly Revenue by Region (USD) Region Jan Feb Mar ----------- --------- -------- -------- North America $45,200 $48,100 $52,300 Europe $31,800 $33,200 $35,900 Asia Pacific $22,400 $24,700 $27,100 Latin America $8,900 $9,400 $10,200 Middle East $5,100 $5,800 $6,300 ----------- --------- -------- -------- Total $113,400 $121,200 $131,800
Dense TextSynthetic
Dense Legal Paragraph
Synthetic dense paragraph — tests small-font continuous text recognition.
▶Ground truth transcript
Section 4.2 - Data Processing Requirements All personally identifiable information (PII) must be encrypted at rest using AES-256 encryption and in transit using TLS 1.3 or higher. Data retention policies require that raw OCR output be purged within 24 hours of processing unless the customer has opted into extended storage (maximum 90 days). Metadata including timestamps, engine IDs, and processing latency may be retained for up to 365 days for analytics purposes. Any cross-border data transfer must comply with GDPR Article 46 requirements.
Dense TextSynthetic
Dense Terms of Service
Synthetic dense paragraph — tests small-font continuous text recognition.
▶Ground truth transcript
Terms of Service - API Usage By accessing the OpenOCR API you agree to the following: (1) You will not exceed your allocated rate limit of 60 requests per minute per API key. (2) You will not use the service to process documents containing protected health information (PHI) without a signed BAA. (3) All API keys are confidential and must not be shared or embedded in client-side code. (4) We reserve the right to throttle or suspend accounts that exhibit abusive patterns. (5) Pricing is subject to change with 30 days written notice. Contact support@openocr.com for questions.
Web UI Screenshot
HabitaOne Website Screenshot
Real website screenshot — tests multilingual UI chrome, small Spanish listing text, and mixed text with graphics.
▶Ground truth transcript
HABITAONE $ 2,000,000 $ 120,000 $ 18,000 $ 47,000 $ 5 $ 250 $ 75,000 $ 330,000 Buy Rent Sell Price Property type Rooms Area State City Save list Show on Map My Properties FOR SALE Terreno en Venta en Puerto Cabello Apartamento en Venta en Los Mangos Terreno en Venta en Capacho Viejo Apartamento en Venta en Res Abadía Puerto Cabello Valencia Carabobo San Diego HabitaOne Linux Claude
See How Engines Perform
Published results use a validated snapshot of this exact OpenOCR Shared 11-Document Corpus (format-normalized scoring).