In plain words: A benchmark of 56 real physics research tasks asks agents to read literature, plan, run experiments, and write a paper, then checks correctness and publishability. The best setup passed just 46 of 168 attempts, with weak reasoning and intuition—not missing knowledge—the main flaw.
Abstract
AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.
Bingran You, Wenjun Ke, Ria Rosenauer, Bo Liu, Wenbo Chen, Min Chen, Shuqi Xu, Yue Ma, Jiechao Feng, Bo Chen, Xiangkai Sun, Jiarui Liu, et al.
arXiv:2611.06910 · cs.AI · submitted Nov 4, 2026
abstract · pdf · html · 41 pages (10 pages main text), 19 figures, 11 tables