MLIP Arena: Advancing Fairness and Transparency in Machine Learning Interatomic Potentials via an Open, Accessible Benchmark Platform

Yuan Chiang (University of California, Berkeley) · Tobias Kreiman (University of California, Berkeley) · Christine Zhang (University of California, Berkeley) · Matthew Kuner (University of California, Berkeley) · Elizabeth Weaver (University of California, Berkeley) · Ishan Amin (University of California, Berkeley) · Hyunsoo Park (Imperial College London) · Yunsung Lim (Seoul National University) · Jihan Kim (Korea Advanced Institute of Science & Technology) · Daryl Chrzan (University of California) · Aron Walsh (Imperial College London) · Samuel Blau (Lawrence Berkeley National Lab) · Mark Asta (University of California, Berkeley) · Aditi Krishnapriyan (UC Berkeley)

benchmark platformchemical reactivitydata leakage density functional theoryerror-based metricsforce field performancemachine learning interatomic potentialsmolecular modeling physical consistencyphysical phenomenaphysics awarenesspredictive accuracy predictive capabilities runtime efficiencythermodynamic propertiestransferability

Machine learning interatomic potentials (MLIPs) have revolutionized molecular and materials modeling, but existing benchmarks suffer from data leakage, limited transferability, and an over-reliance on error-based metrics tied to specific density functional theory (DFT) references. We introduce MLIP Arena, a benchmark platform that evaluates force field performance based on physics awareness, chemical reactivity, stability under extreme conditions, and predictive capabilities for thermodynamic properties and physical phenomena. By moving beyond static DFT references and revealing the important failure modes of current foundation MLIPs in real-world settings, MLIP Arena provides a reproducible framework to guide the next-generation MLIP development toward improved predictive accuracy and runtime efficiency while maintaining physical consistency. The Python package and online leaderboard are available at https://github.com/atomind-ai/mlip-arena.