Reproduce “Flex-RSI: Benchmarking Recursive Self-Improvement of Frontier Agents” 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

Jingdong Zhang, Jionghao Wang, Zhengming Yu, Yancheng Cai, Wei Chow, Chris Choy, Zhiyang Dou, Hang Hua and 14 more

Flex-RSI tests whether agents improve solutions and harnesses. It keeps model weights fixed: an executor revises solutions and an improver edits the harness. On nine graded build tasks, median best-to-worst model gaps grew from 0.9 to 38.8 points; enabling the improver raised test scores in six of seven matched pairs. These are benchmark findings, not weight learning.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.