MatterGen
MicrosoftMatterGen is Microsoft's diffusion model for generating inorganic crystal structures across the periodic table. It ships an unconditional base model and property-conditioned adaptations, along with training data, preprocessing code, a full training pipeline, and evaluation tooling for novelty and predicted stability.
Verified 2026-09-10 via the Hugging Face model card and Microsoft repository.
Openness
5 high confidence- weights
- open(MIT ungated base and property-conditioned checkpoints)
- data
- open(MP-20 and Alex-MP-20 training releases with preprocessing instructions)
- code
- open(full preprocessing, training, fine-tuning, generation, and evaluation pipeline)
- license
- MIT(OSI)
MatterGen publishes the weights, training datasets, and end-to-end code required to train and adapt the model, reaching the full open-source rung.
- https://raw.githubusercontent.com/microsoft/mattergen/main/README.md recorded 2026-09-10
Provides base and fine-tuned checkpoints, downloadable MP-20 and Alex-MP-20 data, preprocessing, training, fine-tuning, generation, and evaluation commands.
- https://huggingface.co/api/models/microsoft/mattergen recorded 2026-09-10
license: mit; gated: false; private: false; base and property-conditioned checkpoints plus data-release files present.
Adoption
2 medium confidenceThe Hub reports zero downloads, so 1,817 GitHub stars are used as a last-resort reach signal.
- https://api.github.com/repos/microsoft/mattergen recorded 2026-09-10
stargazers_count: 1817; archived: false; fork: false.
Capability
3 high confidenceTwo attributes place MatterGen one rung below Prithvi-EO; its unusually complete open pipeline affects openness, not capability.
- https://raw.githubusercontent.com/microsoft/mattergen/main/README.md recorded 2026-09-10
Documents unconditional and multi-property-conditioned crystal generation plus fine-tuning on built-in or user-supplied property data.
Verified 2026-09-10