#OCR
20 related articles
OXR: 1.3B Open-Source Document Parsing Model Tops Three Major Benchmarks
OXR is a 1.3B open-source document parsing model by SoMarkAI that converts PDFs and images to Markdown/JSON, recognizes 13 element types with layout coordinates, and tops three major benchmarks—parsing 100 pages in 101 seconds.
Baidu Open-Sources Unlimited OCR 2: A 3B Model That Parses Hundreds of Pages in a Single Pass
Baidu open-sources Unlimited OCR 2, a 3B MIT-licensed OCR model that uses SR-SWA attention to parse hundreds of pages in one request, scoring 77.1 on MarathonDocBench.
Baidu Releases Unlimited OCR 2: SR-SWA Enables Unlimited-Length Document Parsing
Baidu releases Unlimited OCR 2 with Sparse Reference Sliding Window Attention (SR-SWA), compressing text-visual attention to a fixed reference set for near-constant decoding latency and memory regardless of document length, achieving SOTA on MPDocBench and MarathonDocBench.
Benchpress: Don't Trust OCR Leaderboards—Test With Your Own Documents
Benchpress is an OCR comparison tool that lets you upload one page of your document and test major OCR and document AI models at once, comparing accuracy, speed, and cost.
Local Receipt Recognition in Panama with Gemma for Automated Tax Tables
Open-source project Cuadre uses a local Gemma model to read Panamanian paper receipts, extract RUC and ITBMS fields, and generate tax CSVs. The model only transcribes; code computes. Data never leaves the machine—ideal for accountants and freelancers batching receipts.
Building a Hindi Scanned-Document RAG Assistant: OCR and Hybrid Retrieval in Practice
Developer Abhishek shares lessons building a Hindi/English scanned-circular RAG assistant: OCR quality decides everything, hybrid retrieval and reranking boost results, and citations plus approval workflows make the product trustworthy.
Invoice AI: Snap a Photo, Get Business Intelligence — An Open-Source AI Analyst Arrives
Invoice AI is an open-source project: PaddleOCR extracts invoice data, cross-validation ensures accuracy, and a Gemma 4 assistant answers business questions in natural language for SMBs.
InvoiceSnap AI: Local AI Turns Receipt Photos into Structured Expense Sheets
InvoiceSnap AI is an open-source local receipt scanner combining Tesseract OCR with Gemma via Ollama to extract merchant, date, amount, and category, with editing and CSV export—privacy-focused and low-cost.
How Agentic OCR Tackles Three Million Pages of Messy Scanned Documents
Reducto shares how agentic OCR handles 3 million pages of messy scanned documents behind J-Mail—covering rasterization, extraction schemas, redaction failures, and evaluation challenges.
Strukku: A Receipt Bookkeeping Tool for Freelancers Powered by Open-Weight Models
Strukku is an open-source receipt bookkeeping tool for freelancers, using the Gemma open-weight vision model to extract vendor, date, and amount from photos and export CSV ledgers.
Datalab Open-Sources OmniParseBench: 16,288 OCR Tests to Audit Document Parsers
Datalab releases open-source OCR benchmark OmniParseBench with 16,288 test items across 2,937 pages and 92 languages. Pass rates are sliceable by tag, script, or language, making document parser evaluation auditable and reproducible.
Document Conversion: The Overlooked Productivity Tool and How It Fits Into AI Workflows
Document conversion is overlooked in developer productivity, yet it directly impacts data entry efficiency. This article analyzes OCR's role in developer workflows, the fit of lightweight solutions like LoveOCR, and three principles for integrating image-to-structured-data into automation.
The Data Trap in Screenshots: How OCR Turns Images Back Into Usable Data
Screenshots make saving information easy but using it hard. This article breaks down a complete OCR workflow: from image extraction and format conversion (JSON/CSV/XML/YAML) to result validation, showing why format conversion is the key step to turning screenshot data into usable data.
Verdandi: An Open-Source Document Intelligence Tool at One-Tenth the Cost of Azure
Former EY analyst open-sources Verdandi, a document intelligence platform benchmarked against Azure Content Understanding: 97% vs 96.2% field accuracy, 9.7x lower cost, local OCR, cross-document validation, and human review.
OpenDocRouter Launches: A Unified API Aggregating 130+ Document Parsing Models
OpenDocRouter launches a unified document parsing API aggregating 130+ frontier and open-weight OCR/VLM models, with at-cost pricing, rate limit management, bounding box layout services, and automated routing.
A New Approach to Document Grounding: PDF Geometry + OCR + Visual Evidence Achieves 72.6% Word F1
A developer built a document grounding system using PDF geometry, OCR spatial info, format matching, and visual evidence — achieving 72.6% Word F1 with no external LLMs.
GAD-RL: Improving OCR Transcription Faithfulness with Gated and Attenuated Distillation
GAD-RL uses gated and attenuated on-policy distillation to improve VLM OCR faithfulness, outperforming GRPO by 8.45 points on CHAOS-Bench.
Korean Invoice OCR Model: Deep Learning + Image Preprocessing Achieves 87% F1 Score
An arXiv study proposes a Korean invoice OCR model combining deep learning and image preprocessing, achieving 87% F1 on real invoice data with minimal latency.
How Empty Table Cells Undermine AI Agent Decision Accuracy
Empty table cells cause OCR value misalignment, corrupting AI Agent decision accuracy. Learn why this hidden trap matters and how tools like LlamaParse address it.
Product ReviewslocalOCR: A Locally Deployed Open-Source OCR Solution Supporting Gemma-4 and Other Vision Models
localOCR is an open-source local OCR tool powered by vision language models like Gemma-4 and Llama 3.2. Learn about its architecture, multi-model support, use cases, and advantages over traditional OCR.