Reproduce “LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering” 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

Yi Wang, Haopeng Zhang, Chengxiang Huang, Rui Dai, Kaikui Liu, Piotr Koniusz, Xiangxiang Chu

LoopArena separates runtime control from coding execution by having a Controller guide a separate Worker through structured progress reports, while holding the Worker and execution setup fixed across comparisons. On 80 full tasks, GPT-5.5 achieves a 36.25% Strict Success Rate, versus 31.25% for fixed control; results remain difficult, and the comparison does not isolate the cause.

Summary by alphaXiv.

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