General-purpose foundation models surpass specialized ones on chemical property prediction tasks
Abstract
Foundation models (FMs) learn reusable representations from large collections of unlabeled data. In chemistry the corresponding data mainly consists of molecular structures, which are abundant, whereas experimental measurements are scarce. Chemical FMs such as ChemBERTa and MoLFormer were therefore trained on diverse molecular structures to enable fine-tuning on small experimental datasets (less than 100 datapoints in most realistic scenarios) and subsequent property prediction. At the same time, general-purpose FMs such as GPT-6 Astra and Claude Opus 5.5 were trained on virtually all publicly available data, including knowledge of the humankind on molecular properties and laws governing them. Here we compare accuracies of chemical and general-purpose FMs on tasks from various areas of chemistry. Strikingly, general-purpose GPT-6 Astra outperforms the currently eligible chemical FMs on five physicochemical and quantum-chemical regression tasks when only 60 experimental datapoints are available. Only when provided 10x datapoints do the best chemical FMs reach performance of GPT-6 Astra: on 1,000 and 10,000 points they show performance on par with Astra on 60 and 1,000 points, respectively. Thus, counterintuitively, frontier general-purpose FMs are better suited for guiding chemical research in data-scarce areas than specialized chemical FMs, suggesting that learning relevant primitives (e.g. chemical structures) does not substitute training on relevant humankind knowledge.
Related
- Accelerating Mechanistic Understanding and Prediction of Chemical Selectivity with Foundation Machine-Learned Interatomic Potentials
- Materials Property Prediction with a Site-Resolved Three-Dimensional Voxel Convolution Neural Network
- Accelerating discovery of phosphorescent blue OLED emitters using chemical foundation models
- Closed-loop Auto Research for Molecular Property Prediction: Discovering and Certifying Generalizable Improvements
- QALPA: Property-guided diffusion modeling for efficient exploration of chemical spaces of flexible molecules
BibTeX
@misc{ostarkov2026generalpurpose,
title = {{General-purpose foundation models surpass specialized ones on chemical property prediction tasks}},
author = {Stepan N. Ostarkov and Gregorii A. Semakin and Ivan A. Bespalov and Michael G. Medvedev},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.41899},
eprint = {2610.41899},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.41899},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - General-purpose foundation models surpass specialized ones on chemical property prediction tasks
AU - Stepan N. Ostarkov
AU - Gregorii A. Semakin
AU - Ivan A. Bespalov
AU - Michael G. Medvedev
PY - 2026
DA - 2026/10/30
PB - arXiv
DO - 10.48550/arxiv.2610.41899
UR - https://arxiv.org/abs/2610.41899
AB - Foundation models (FMs) learn reusable representations from large collections of unlabeled data. In chemistry the corresponding data mainly consists of molecular structures, which are abundant, whereas experimental measurements are scarce. Chemical FMs such as ChemBERTa and MoLFormer were therefore trained on diverse molecular structures to enable fine-tuning on small experimental datasets (less than 100 datapoints in most realistic scenarios) and subsequent property prediction. At the same time, general-purpose FMs such as GPT-6 Astra and Claude Opus 5.5 were trained on virtually all publicly available data, including knowledge of the humankind on molecular properties and laws governing them. Here we compare accuracies of chemical and general-purpose FMs on tasks from various areas of chemistry. Strikingly, general-purpose GPT-6 Astra outperforms the currently eligible chemical FMs on five physicochemical and quantum-chemical regression tasks when only 60 experimental datapoints are available. Only when provided 10x datapoints do the best chemical FMs reach performance of GPT-6 Astra: on 1,000 and 10,000 points they show performance on par with Astra on 60 and 1,000 points, respectively. Thus, counterintuitively, frontier general-purpose FMs are better suited for guiding chemical research in data-scarce areas than specialized chemical FMs, suggesting that learning relevant primitives (e.g. chemical structures) does not substitute training on relevant humankind knowledge.
ER -