Reproduce “Evaluation-Free Recursive Self-Improvement for Coding 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
Zhe Ren, Guowei Rong, Yimeng Chen, Nanbo Li, Piotr Piękos, Mateusz Ostaszewski, Dandan Guo, Jürgen Schmidhuber
ELITE improves coding agents by integrating compatible mechanisms from independently implemented versions of a shared scaffold, without using candidate-task outcomes to guide evolution. On a fixed 60-task SWE-bench Pro subset, DeepSeek-V4-Flash success rises from 60.0% to 85.6%; after evolution, scores are 73.3% on Verified and 69.3% on Polyglot. These results concern selected coding-agent benchmarks and do not establish broader generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.