DR Pixel: open-source research agent that searches page images without OCR

Tiansheng Hu introduces DR Pixel, an open-source deep research agent that searches and reasons directly over page images without OCR, according to an X post on November 3, 2026.

The work is joint research from BerkeleyNLP and rtnldynamics, with the paper listed on arXiv as 2611.03097.

Hu says that on 1.06M pages, the authors' trained 27B model matches GPT-5.4. The claim comes from the authors' post, and this report does not include independent evaluation.