Open OCR models update: Chandra and OlmOCR-2 added; guidance for open-weight OCR options

AI Impact Summary

Open OCR models update introduces Chandra and OlmOCR-2 and highlights the cost, privacy, and performance benefits of open-weight OCR for document pipelines. The guide clarifies when to fine-tune vs use models out-of-the-box and stresses layout-aware outputs with grounding to preserve reading order and reduce hallucination. Downstream systems must align on output formats (DocTags, HTML, Markdown, JSON) and consider prompts, task switching, and multimodal capabilities to extend OCR into document QA and retrieval.