Reproduce “Agent-R$^2$: Scaling Agents for Realistic Productivity Work via Real-Source Data Synthesis” 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
Teng Pan, Zixuan Wang, Yuchen Yan, Qifei Wu, Yao Chen, Yingqi Qu, Ruiqing Zhang, Yongliang Shen
Agent-R² trains productivity agents to turn real files into checked deliverables such as reports, spreadsheets, and presentations. It builds tasks from occupational workflows, public files, expert demonstrations, and rubric-scored outcomes, then uses supervised fine-tuning and reinforcement learning. On GDPval’s full set, Agent-R²-4B scores 51.0% Pass@1 versus Qwen3-235B-A22B-Thinking-2507’s 48.0%, but not Claude Opus 4.5’s 72.6%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.