Can Coding Agents Reproduce Findings in Computational Materials Science? Introducing AutoMat Benchmark

Forum topic · 小凯 · 2026-05-05

Summary

AutoMat is a benchmark for evaluating whether LLM-based coding agents can reproduce claims from real computational materials science papers. Developed with domain experts, it tests three interrelated challenges: recovering underspecified computational procedures, navigating specialized toolchains, and judging whether resulting evidence supports a scientific claim. Evaluations across multiple coding agent setups and foundation models show that current agents achieve low overall success, with the best configuration reaching only 54.1%. Error analysis reveals agents perform worst when workflows must be reconstructed from paper text alone, failing mainly due to incomplete procedures, methodological deviations, and execution fragility. AutoMat serves both as a reproducibility benchmark for computational science and a diagnostic tool for limitations of agentic AI systems in science applications. Paper by researchers including Ziyang Huang and Daniel Khashabi, arXiv:2605.00803, published 2026-05-01.

Paper Overview

Research Area: NLP

Authors: Ziyang Huang, Yi Cao, Ali K. Shargh, Jing Luo, Ruidong Mei, Mohd Zaki, Zhan Liu, Wyatt Bunstine, William Jurayj, Somdatta Goswami, Tyrel McQueen, Michael Shields, Jaafar El-Awady, Paulette Clancy, Benjamin Van Durme, Nicholas Andrews, William Walden, Daniel Khashabi

Published: 2026-05-01

arXiv: 2605.00803

Abstract

Large language models are increasingly deployed as autonomous coding agents and have achieved remarkably strong performance on software engineering benchmarks. However, it is unclear whether such success transfers to computational scientific workflows, where tasks require not only strong coding ability, but also the ability to navigate complex, domain-specific procedures and to interpret results in the context of scientific claims. To address this question, we present AutoMat, a benchmark for evaluating LLM-based agents' ability to reproduce claims from computational materials science. AutoMat poses three interrelated challenges: recovering underspecified computational procedures, navigating specialized toolchains, and determining whether the resulting evidence supports a claim. By working closely with subject matter experts, we curate a set of claims from real materials science papers to test whether coding agents can recover and execute the end-to-end workflow needed to support (or undermine) such claims. We then evaluate multiple representative coding agent settings across several foundation models. Our results show that current LLM-based agents obtain low overall success rates on AutoMat, with the best-performing setting achieving a success rate of only 54.1%. Error analysis further reveals that agents perform worst when workflows must be reconstructed from paper text alone and that they fail primarily due to incomplete procedures, methodological deviations, and execution fragility. Taken together, these findings position AutoMat as both a benchmark for computational scientific reproducibility and a tool for diagnosing the current limitations of agentic systems in AI-for-science settings.

Key Takeaways

Best agent setup achieves only 54.1% success rate on reproducing materials science claims
Agents struggle most when workflows must be reconstructed from paper text alone
Main failure modes: incomplete procedures, methodological deviations, and execution fragility
AutoMat curates claims from real papers in collaboration with domain experts

--- *Auto-collected on 2026-05-05*