OCR Examples Gallery

Isaac FlathIsaac FlathAugust 11, 2026

A comparison of OCR tools and vision-language models on document pages: dense tables, forms, charts, maps, scanned equations, and diagrams. Every model sees the same page and (for VLMs) the same doc-type prompt.

The six models represent six ways someone might build a production document system. This is a curated comparison rather than an exhaustive model ranking:

ApproachRepresentativeWhy it is hereSimilar models
Conventional OCRPaddleOCR PP-OCRv5A direct text-recognition engine without a document or language-model layerRapidOCR, Tesseract, EasyOCR
Managed document OCRAWS TextractThe baseline for a team that wants to buy an API rather than operate OCR infrastructureGoogle Document AI, Azure AI Document Intelligence, ABBYY Vantage
Document pipelineMinerUA packaged system that combines recognition, layout, and document reconstructionDocling, Unstructured, Marker
OCR-specific vision modelChandra OCR 2A vision model built specifically for document transcriptionSurya OCR 2, Nanonets-OCR2-3B, DeepSeek-OCR
Open general VLMQwen3-VL 32BA self-hostable general vision-language model prompted for extractionQwen3-VL 8B, InternVL, Molmo
Hosted general VLMGemini 3.5 FlashA general multimodal model consumed as a managed APIOpenAI GPT, Anthropic Claude, Gemini Pro

Models in the last column cover the same broad role. This gallery should help you pick a general class, then you need to evaluate models on your own pdfs to narrow it down further.

This is not meant to be an instructive guide, for that start with these 2 articles:

  1. How to Choose an OCR Model
  2. VLMs for Hard OCR Tasks

Hope you enjoy!

Government financial statement

Source: Financial Report of the United States Government, FY2024, page 68 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

1950 census questionnaire grid

Source: 1950 Census population questionnaire, page 1 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Coastal chart with coordinates

Source: NOAA United States Coast Pilot 2, page 240 (page PDF) · doc type: map

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

United States v. Banque Pictet & Cie SA — Deferred Prosecution Agreement

Source: United States v. Banque Pictet & Cie SA — Deferred Prosecution Agreement, page 12 (page PDF) · doc type: generic

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

US 11,335,153 B1 — Fully Autonomous System for Grading, Stamping and Encapsulating a Collectible Object

Source: US 11,335,153 B1 — Fully Autonomous System for Grading, Stamping and Encapsulating a Collectible Object, page 5 (page PDF) · doc type: diagram

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Alameda County — Official Retail Food Inspection Report

Source: Alameda County — Official Retail Food Inspection Report, page 1 (page PDF) · doc type: form

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Financial Accounts of the United States: Second Quarter 2025 (Z.1)

Source: Financial Accounts of the United States: Second Quarter 2025 (Z.1), page 84 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Berkshire Hathaway Inc. 2024 Annual Report

Source: Berkshire Hathaway Inc. 2024 Annual Report, page 90 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

FDIC Quarterly Banking Profile, Fourth Quarter 2025

Source: FDIC Quarterly Banking Profile, Fourth Quarter 2025, page 11 (page PDF) · doc type: chart

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Apple Inc. Form 10-K, fiscal 2024

Source: Apple Inc. Form 10-K, fiscal 2024, page 34 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Health Insurance Claim Form — CMS-1500 (02-12)

Source: Health Insurance Claim Form — CMS-1500 (02-12), page 1 (page PDF) · doc type: form

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

GOPRELTO (cocaine hydrochloride) Nasal Solution — Prescribing Information

Source: GOPRELTO (cocaine hydrochloride) Nasal Solution — Prescribing Information, page 3 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

RITALIN and RITALIN-SR (methylphenidate hydrochloride) — Prescribing Information

Source: RITALIN and RITALIN-SR (methylphenidate hydrochloride) — Prescribing Information, page 1 (page PDF) · doc type: generic

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Fictional discharge summary, 'John Doe' (GeriSage teaching example)

Source: Fictional discharge summary, 'John Doe' (GeriSage teaching example), page 1 (page PDF) · doc type: generic

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Cheney Brothers, 'Art in Fashion' typescript draft (Library of Congress)

Source: Cheney Brothers, 'Art in Fashion' typescript draft (Library of Congress), page 43 (page PDF) · doc type: generic

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Calvin Coolidge, speech before the American Classical Association, typescript (Library of Congress)

Source: Calvin Coolidge, speech before the American Classical Association, typescript (Library of Congress), page 21 (page PDF) · doc type: generic

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Attention Is All You Need (arXiv:1706.03762)

Source: Attention Is All You Need (arXiv:1706.03762), page 4 (page PDF) · doc type: equations

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

First Bus Glasgow — Service 4/4A Timetable

Source: First Bus Glasgow — Service 4/4A Timetable, page 1 (page PDF) · doc type: table

AWS Textract AnalyzeDocument (boto3 1.43.67); no task prompt.

Vanguard S&P 500 Index ETF (VFV) fact sheet

Source: Vanguard S&P 500 Index ETF (VFV) fact sheet, page 2 (page PDF) · doc type: table

chandra-ocr 0.2.0 (Chandra OCR 2, hf method); no task prompt.