Researchers have created a new benchmark called LongDocBench to evaluate the ability of AI systems to understand and reconstruct the structure of long documents. The benchmark includes 85 real-world documents with human-verified annotations for table-of-contents hierarchies and contextual relationships between objects such as tables, figures, and captions. The goal is to improve document-level structure recovery in long documents, which is essential for applications like question-answering and information retrieval. --- Why it matters: This matters because many real-world documents are long and complex, requiring AI systems to understand their structure and relationships between objects to provide accurate answers or extract relevant information. LongDocBench provides a standardized way to evaluate the performance of AI systems on these tasks, which can lead to improvements in document intelligence and related applications. Source: https://arxiv.org/abs/2608.15064