AI agents hold growing promise for accelerating scientific discovery; yet, a lack of frontier evaluations hinders adoption into real workflows. Expert-written benchmarks have proven effective at measuring AI reasoning, but most at this stage have become saturated and only measure performance on constrained outputs. To help address this gap, we introduce COMPOSITE-STEM, a benchmark of 70 expert-written tasks in physics, biology, chemistry, and mathematics, curated by doctoral-level researchers. Our benchmark combines exact-match grading and criterion-based rubrics with an LLM-as-a-jury grading protocol, allowing more flexible assessment of scientifically meaningful outputs. Using an adapted multimodal Terminus-2 agent harness within the Harbor agentic evaluation framework, we evaluate four frontier models. The top-performing model achieves 21%, demonstrating that COMPOSITE-STEM captures capabilities beyond current agent reach. All tasks are open-sourced with contributor permission to support reproducibility and to promote additional research towards AI's acceleration of scientific progress in these domains.
@article{arxiv.2604.09836,
title = {COMPOSITE-Stem},
author = {Kyle Waters and Lucas Nuzzi and Tadhg Looram and Alessandro Tomasiello and Ariel Ghislain Kemogne Kamdoum and Bikun Li and Damien Sileo and Egor Kretov and Francesco Fournier-Facio and Georgios Soloupis and Haile Kassahun and Hew Wolff and Jiaqi Cai and Lianghui Li and Marc Roth and Mohinder Naiya and Naixu Guo and Qicheng Tang and Richard Wheeler and Samuele Sala and Serguei Popov and Steven Dillmann and Yuqi Li},
journal= {arXiv preprint arXiv:2604.09836},
year = {2026}
}