Reproduce “LiteEvo: Automated, Cost-Efficient Harness Evolution for Generalization to Unseen 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
Euntae Choi, Sumin Song, Sungjoo Yoo
LiteEvo automatically improves an agent’s instructions and action-handling setup without changing model weights, using attempts on training tasks to revise a harness. With frozen Qwen3.5-9B, its evolved harness raises AppWorld test success from zero to 65.5% within two attempts. This is evidence on unseen tasks within AppWorld, not transfer to unseen benchmarks; API-cost results exclude local inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.