Reproduce “DR Pixel: Page-Native Deep Research with Vision-Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tiansheng Hu, Yichuan Wang, James V. Talwar, Nicholas Tomlin, Chen Zhao, Sewon Min
DR Pixel keeps scientific-paper pages as images during retrieval and reasoning, allowing an agent to search, inspect visual evidence, and search again. On DRPixelBench’s 500 questions, GPT-5.4 reaches 51.9% answer accuracy with 9.2 retrieval calls per question, versus 46.3% and 13.8 calls for OCR text-plus-figure search. These results concern scientific-paper evaluation, not elapsed-time efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.