The field of Optical Character Recognition (OCR) and related document processing technologies is witnessing significant advancements, particularly in the integration of large language models (LLMs) and vision-language models (VLMs). Recent developments highlight the growing sophistication in handling complex, multilingual, and multi-modal document processing tasks. Innovations are being driven by the need to improve accuracy, reduce uncertainty, and enhance the robustness of OCR systems, especially in scenarios involving noisy or degraded documents. Additionally, there is a notable shift towards leveraging model uncertainty metrics to guide the processing of documents, ensuring more reliable and accurate information extraction. The integration of VLMs and LLMs is proving to be particularly effective in tasks requiring deep semantic understanding and context-aware processing, such as in the deciphering of ancient scripts and the generation of LaTeX code from handwritten math expressions. Furthermore, the field is increasingly focusing on benchmarking and evaluating these models to identify strengths and weaknesses, driving further advancements in model capabilities and performance. Notably, the development of comprehensive benchmarks is playing a crucial role in this evolution, providing standardized tests for evaluating the literate capabilities of multimodal models in diverse and challenging scenarios.