DR Pixel Enables Page-Native Research Using Vision-Language Models
November 1, 2026
DR Pixel is an open-source multimodal agent that performs deep research by reasoning directly over rendered document page images rather than OCR text. It outperforms OCR-based baselines on the DRPixelBench benchmark, which requires reasoning over tables, figures, and equations.
HOW THIS AFFECTS YOU
●
builderYou can build more accurate research tools by using rendered images instead of noisy OCR text.
●
researcherThis establishes a new multimodal benchmark for document-based reasoning tasks.