June 29, 2026
Sign up, get your API key, and extract tables from your first PDF. Complete walkthrough with curl, Python, and JavaScript. Covers output formats, OCR language settings, and best practices for accurate extraction.
Start by getting your free API key at pdfxdoc.com/signup. No credit card required.
Step 1: Get your key. Sign up and copy your doc_... API key. Store it in an environment variable: export PDFXDOC_KEY=doc_your_key_here
Step 2: Send a PDF. Use curl with your key:
curl -X POST https://pdfxdoc.com/api/v1/convert \
-H "X-Api-Key: $PDFXDOC_KEY" \
-F "[email protected]" \
-F "output_format=json"
Step 3: Get your result. Poll the result endpoint with the task ID:
curl -H "X-Api-Key: $PDFXDOC_KEY" \
https://pdfxdoc.com/api/v1/result/TASK_ID
Output formats: json (structured), docx (Word), md (Markdown), html (web). Set ocr_lang=ara,eng for Arabic + English documents. For tables with complex layouts, use table_mode=accurate.
Full API reference: docs.
GuideGetting Started
June 28, 2026
Deep dive into processing Arabic and mixed Arabic-English documents. Configuring OCR parameters, handling RTL text, and optimizing table detection for Arabic invoices. Benchmarks and accuracy tips.
pdfxdoc is the only PDF extraction API that defaults to Arabic + English OCR. Here's how to get the best results.
Language codes: Set ocr_lang=ara,eng for bilingual documents. The comma-separated list tells the OCR engine which character sets to expect. For pure Arabic: ocr_lang=ara. For Arabic + French: ocr_lang=ara,fra.
RTL text handling: JSON output preserves logical character order. When rendering, set direction: rtl in CSS for Arabic text blocks. DOCX output applies RTL paragraph direction automatically.
Table detection: Arabic invoices often use right-to-left column layouts. Set table_mode=accurate for the best detection. The default fast mode works well for English tables but can misalign Arabic columns.
Accuracy tips: Use PDFs with embedded text (not scanned images) for near-perfect accuracy. For scanned documents, ensure at least 200 DPI resolution. Mixed Arabic-English documents average 96%+ character accuracy on clean PDFs.
ArabicOCR
June 27, 2026
We benchmarked Docling against the top PDF extraction APIs on accuracy, speed, pricing, and data privacy. How we compare on table extraction, Arabic OCR, and total cost of ownership.
We tested all three APIs on a benchmark of 100 documents: invoices, reports, academic papers, and forms split 50/50 English and Arabic.
Accuracy (table extraction): pdfxdoc: 94.2%, LlamaParse: 95.1%, Mistral OCR: 92.8%. LlamaParse edges ahead on complex tables, but the gap narrows significantly with table_mode=accurate.
Arabic OCR: pdfxdoc: 96.1%, Mistral OCR: 91.3%, LlamaParse: 88.7%. pdfxdoc's default Arabic+English model gives a clear advantage for bilingual documents.
Speed (avg 5-page PDF): pdfxdoc: 3.2s, Mistral OCR: 4.1s, LlamaParse: 5.8s. pdfxdoc benefits from running on dedicated hardware without shared cloud queues.
Pricing (5,000 pages/month): pdfxdoc: $9/mo (Pro), Mistral OCR: ~$30/mo (PAYG), LlamaParse: $75/mo (Business). pdfxdoc is the most cost-effective at scale.
Data privacy: pdfxdoc processes on private infrastructure. No data ever touches third-party cloud AI services. Both competitors use cloud-hosted models where your PDFs leave your control.
Bottom line: pdfxdoc wins on Arabic OCR, speed, pricing, and privacy. LlamaParse wins on table accuracy for very complex layouts. Mistral OCR is a solid generalist but doesn't lead in any category.
ComparisonBenchmark