An Analysis of Falling Costs and Optimization Strategies in Open OCR Models

An analysis of eight open OCR models run across 850 pages shows that costs have dropped significantly over the past 18 months, driven largely by new models like LightOnOCR-3-0.8B. The author found that optimization techniques such as running models at 2,048 pixels rather than full-size scans reduce costs without sacrificing accuracy on book pages. Open article →