The ParseBench document parsing capability evaluation dataset was released by the LlamaIndex team in 2024–2025, and the related paper results are as follows: ParseBench: A Document Parsing Benchmark for AI AgentsIt aims to promote the evolution of document parsing from traditional OCR to structured understanding, and support the evaluation and optimization of multimodal models and information extraction systems. This dataset contains approximately 2,000 manually validated and labeled pages and 169,011 test rules across five dimensions. These pages are taken from publicly available corporate documents covering insurance, finance, government, and other sectors, encompassing various page types including PDFs, scanned images, and pages containing tables and layout structures. Standardized parsing results are provided and aligned with human annotations to evaluate the model's performance in structural understanding and information extraction.