Zatom-1

This repository contains the model weights for Zatom-1, the first end-to-end foundation model that unifies generative and predictive learning of 3D molecules and materials. Introduced in Zatom-1: A Multimodal Flow Foundation Model for 3D Molecules and Materials, Zatom-1 is a Transformer trained with a multimodal flow matching objective that jointly models discrete atom types and continuous 3D geometries.

GitHub repository

For the full implementation, training scripts, and configuration files, visit: https://github.com/Zatom-AI/zatom

Sample Usage

Installation

To get started, clone the repository and install the dependencies:

# Clone project
git clone https://github.com/Zatom-AI/zatom
cd zatom

# Install requirements
pip install -e '.[cuda]'

Evaluation

To generate evaluation metrics for molecule and material generation using the Zatom-1 weights:

python zatom/eval_fm.py \
    ckpt_path=checkpoints/zatom_1_joint_paper_weights.ckpt \
    model.sampling.num_samples=10000 \
    model.sampling.batch_size=1000 \
    name=eval_run \
    seed=42 \
    trainer=gpu

Open-source resources

Zatom-1 builds upon the source code and data from the following projects:

• all-atom-diffusion-transformer
• flow_matching
• flowmm
• jvp_flash_attention
• lemat-genbench
• lightning-hydra-template
• PlatonicTransformers
• ProteinWorkshop
• posebusters
• tabasco

We thank all their contributors and maintainers!

Acknowledgements

This research used resources of the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 using AI4Sci@NERSC award NERSC DDR-ERCAP0036206 awarded to AM. NBE would like to acknowledge support from the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research, EXPRESS: 2025 Exploratory Research for Extreme-Scale Science program, and the Scientific Discovery through Advanced Computing (SciDAC) program, under Contract Number DE-AC02-05CH11231 at Berkeley Lab.

Citation

If you use the code or data associated with this package or otherwise find this work useful, please cite:

@article{zatom_1_2026,
    title={Zatom-1: A Multimodal Flow Foundation Model for 3D Molecules and Materials},
    author={Alex Morehead* and Miruna Cretu* and Antonia Panescu* and Rishabh Anand* and Maurice Weiler* and Tynan Perez* and Samuel Blau and Steven Farrell and Wahid Bhimji and Anubhav Jain and Hrushikesh Sahasrabuddhe and Pietro Liò and Tommi Jaakkola and Rafael Gómez-Bombarelli and Rex Ying* and Ben Erichson* and Michael Mahoney*},
    year={2026},
    eprint={2602.22251},
    archivePrefix={arXiv},
    primaryClass={cs.LG},
    url={https://arxiv.org/abs/2602.22251},
    note={* denotes equal contribution}
}