Gemini 3 Pro marks a significant advancement in vision AI, particularly in the challenging domain of document understanding. Real-world documents are inherently complex—often unstructured, cluttered, and inconsistent, featuring a mix of images, handwritten notes, nested tables, intricate mathematical expressions, and non-linear layouts. Gemini 3 Pro delivers a transformative improvement across the full document processing pipeline, combining state-of-the-art optical character recognition (OCR) with advanced visual reasoning to handle these challenges with remarkable precision. At the core of its capabilities is intelligent perception—enabling the model to accurately detect and interpret text, tables, mathematical formulas, figures, and charts, even when they appear in noisy or unconventional formats. This includes robust performance on low-quality scans, faded handwriting, and documents with overlapping elements. A key innovation is the model’s ability to perform "derendering"—the process of reconstructing a visual document into its underlying structured code, such as HTML, LaTeX, or Markdown, that can faithfully reproduce the original layout and content. This capability allows for seamless conversion of complex, multi-layered documents into editable, machine-readable formats. For example, Gemini 3 Pro can accurately convert an 18th-century merchant’s handwritten logbook into a well-structured table, preserving both content and context. In another case, it can analyze a scanned image containing mathematical annotations and generate precise LaTeX code that captures every symbol, equation, and formatting detail. These feats demonstrate not just recognition, but true understanding of document structure and meaning. By bridging the gap between visual input and structured output, Gemini 3 Pro sets a new benchmark for AI-driven document intelligence—unlocking powerful applications in legal, scientific, financial, and archival domains where accurate, automated document processing is essential.