Open-source DR Pixel agent does OCR-free deep research over page images, 27B model matches GPT-5.4
CShorten30 · x · 2026-11-04
Berkeley NLP and rtnldynamics released DR Pixel, billed as the first open-source deep research agent that searches and reasons directly over page images without OCR. The team indexed recent CS and biology papers for cross-domain deep research; DR Pixel reads figures, tables, and equations directly and beats OCR-based search on their benchmarks. A trained 27B model on 1.06M pages reportedly matches GPT-5.4. Led by Hughie Hu, it's a follow-up to PixelRAG.
Related event: Berkeley and Partners Open-Source DR Pixel, an OCR-Free Deep Research Agent(5 posts)→
More from coding & agent
- CMU Study: Harness Engineering Boosts Scientific Agents Without Retraining — niloofar_mire · 2026-11-04
- OpenClaw WebUI Update Adds Parallel Task Panel for Running Multiple Agents — heyneighbor · 2026-11-04
- OpenAI ships Skills Over MCP while the spec (SEP-2640) remains unmerged — tristanbob · 2026-11-04
- Anaconda on deslopping AI prose: the structural tells in PR descriptions — anacondainc · 2026-11-04
- Source-code study of 11 coding agents finds every harness shares the same 7 parts — blaizedsouza · 2026-11-04
- UHP open protocol standardizes agent harnesses: run Codex, Claude Code via one API — blaizedsouza · 2026-11-04