Insanely great OCR

Datalab's open-source Marker and Chandra convert images and PDFs into Markdown, HTML, JSON, and other structured formats with astonishing accuracy, even on degraded text. Their hosted service runs an improved version of Chandra that promises higher accuracy than the open weights. It includes $20 in free monthly credits, zero data retention by default, SOC 2 Type 2 compliance, and custom BAAs. Opting into data retention earns a 25% discount (sensitive files are best kept off cloud services in any case). Pricing per 1,000 pages: $4 for Fast or Balanced mode; $10 for Accurate. Tested with 1,525 pages of poorly typed and copied material through Balanced mode, with astounding results. A few sample pages: Original PDF | Markdown. The whole job took just a few minutes and used only $4.57 of the $20 in free monthly credits. Kick the tires in the Playground.