NatureBench: Can Coding Agents Match the Published SOTA of Nature-Family Papers?
For researchers evaluating AI coding agents on real scientific problems, this benchmark reveals that current agents lack genuine scientific invention and mostly succeed through methodological translation.
The paper introduces NatureBench, a benchmark of 90 tasks from Nature-family papers, and finds that the best AI coding agent surpasses published SOTA on only 17.8% of tasks, with failures mainly due to wrong method choice and insufficient compute.
We introduce NatureBench, a cross-discipline benchmark of 90 tasks distilled from peer-reviewed Nature-family publications, designed to evaluate whether AI coding agents can move beyond reproduction toward discovery on real scientific problems. NatureBench is built on NatureGym, an automated pipeline that constructs a standardized, per-task containerized environment from a source paper, addressing the environment-fragmentation problem that has limited the credibility of prior agent-on-research benchmarks. Evaluating ten frontier agent configurations under a strict web-search-disabled protocol, we find that the strongest model surpasses SOTA on only 17.8% of tasks under the g>0.1 criterion. Analysis of method pathways reveals that agents succeed primarily through methodological translation, converting scientific tasks into familiar supervised prediction problems, rather than through genuine scientific invention. Failures are dominated by wrong method choice and insufficient compute budget, not by task misunderstanding. We release the benchmark, the NatureGym pipeline, and a public leaderboard with maintainer-side reproduction. Code: https://github.com/FrontisAI/NatureBench