Chandra OCR 2 Beats GPT-5 Mini and Gemini on Handwriting Recognition — 5.7K Stars, +546/Day

GitHub·medium signal

Datalab's Chandra OCR 2 is trending at 5,761 stars (+546/day) as an open-source OCR model that converts images and PDFs into structured HTML/Markdown/JSON with full layout preservation. On a 43-language benchmark, Chandra 2 scored 77.8% average accuracy vs Gemini 2.5 Flash at 67.6% and GPT-5 Mini at 60.5%. Supports 90+ languages, complex tables, math equations, and handwriting with both local (HuggingFace) and remote (vLLM) inference modes.