Reproduce “EvoCowork: A Benchmark for Real-World Office Tasks” 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
Zhou Liu, Xingyan Chen, ZhaoYang Han, Zewei Pan, Zeli Su, Ningxin Pan, Xue Hu, Kaixiong Ting and 1 more
EvoCowork tests whether harness evolution transfers from single-modality tasks to coordinated cross-office work while model weights remain fixed. Its benchmark scores submitted artifacts and holds cross-office tasks out of feedback and selection. With GLM 5.3 and the frozen MSA harness, ACE and AHE improve in-domain scores by 4.90 and 2.64 points, but cross-office changes are +0.03 and −1.72 in one-run comparisons.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.