DR Pixel: open-source 27B deep research agent reads paper pages without OCR, matches GPT-5.4
CShorten30 · x · 2026-11-04
A joint team from UC Berkeley NLP and Rtnldynamics released DR Pixel, described as the first open-source deep research agent that searches and reasons directly over page images — no OCR required. It reads figures, tables, and equations natively.
- They indexed 1.06M recent CS and biology pages so users can run cross-domain deep research in one place
- Their trained 27B model matches GPT-5.4 on their benchmarks and beats OCR-based search
- A follow-up to PixelRAG, led by @HughieHu43; code and paper are open
Related event: Berkeley and Partners Open-Source DR Pixel, a 27B Deep Research Agent(4 posts)→
More from coding & agent
- Unity ships official Codex plugin with skills, four demo scenarios — OpenAIDevs · 2026-11-04
- Dev quips: a 6-person team's project is just one Claude Code prompt — burny_tech · 2026-11-04
- Dev Builds a Voice-Only Canvas Workspace Accessible From His Phone — NERDDISCO · 2026-11-04
- LangChain's Chase: Browsers Are Becoming the New Terminal for AI Agents — BraceSproul · 2026-11-04
- Google Cloud ships Antigravity plugin with guardrails and live docs grounding — rseroter · 2026-11-04
- Celesto CLI Lets AI Agents Inspect CI Jobs and Logs — aniketmaurya · 2026-11-04