Reproduce “Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?” 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
Ming Zhong, Stacy Li, Nicholas Carlini, Matthew Jagielski
Agent4Sci evaluates whether coding agents can improve verified scientific models in research codebases under fixed compute budgets. It standardizes 20 benchmarks and separates agent work from held-out testing and audits. At least one frontier agent improved the reproduced baseline on every benchmark; Claude Opus 4.6 achieved a 10.4% mean relative improvement. Most gains reflected familiar ML engineering, not demonstrated research-level insight.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.