OCR Benchmark

The benchmark evaluates OCR accuracy and performance between traditional OCR providers and Vision Language Models (VLMs) using various real-world documents, including messy scans. It uses open-source datasets and methodologies, with results showing VLMs often matching or exceeding traditional OCR in certain scenarios like low-quality scans and handwritten documents. Traditional models may perform better on high-density text pages. The results include measurements of accuracy, cost, and latency.