AI Potluck
Back to Gap Map Model components / Scientific AI models

MatterSim

Microsoft
open weights / Overall score: 1.7

MatterSim is Microsoft's deep-learning atomistic model for bulk materials, acting as a machine-learning force field across the periodic table for temperatures from 0 to 5000 K and pressures up to 1000 GPa. It ships 1M- and 5M-parameter M3GNet checkpoints in the repository, with a fine-tuning script for adapting the model to a custom dataset or a higher level of theory.

The pretrained checkpoints ship inside the repository and through the mattersim PyPI package; more advanced versions are held back behind Azure Quantum Elements. Verified 2026-09-10 via the GitHub API, the repository README, and the model card.

Openness

3 high confidence
3.0
weights
open(MIT MatterSim-v1.0.0-1M and MatterSim-v1.0.0-5M checkpoints in the repository)
data
closed(the 3M and 6M-structure training sets are not released
code
partial(inference and fine-tuning scripts
license
MIT(OSI)

The released v1 checkpoints are MIT and locally runnable, but the training data stays inside Microsoft and the repository documents fine-tuning rather than a pretraining run.

Adoption

1 medium confidence
1.0

No Hugging Face artifact is declared, so 594 GitHub stars are the last-resort signal; the mattersim PyPI package is the channel that actually ships the weights and is not banded here.

Capability

2 high confidence
2.0

One attribute. The model card rules out generation explicitly, the temperature and pressure ranges are conditions rather than scales, and the 1M and 5M checkpoints are sizes.

Verified 2026-09-10