Benefit Ledger · verified · Materials Science
Microsoft’s MatterGen diffusion model, published in Nature in January 2025, more than doubled the rate of stable, unique, novel generated crystals versus prior generative models and reported one synthesized material within 20 percent of a target property.
Current score
+1.74
Major generative-materials method with limited synthesis (tier 3). High attribution. Nature. Realization is mostly computational plus one experiment. Medium durability as models improve.
MatterGen jointly denoises atom types, coordinates, and lattices and can be fine-tuned toward chemistry, space group, or scalar properties. The paper’s headline experimental proof of concept is a single synthesized structure (reported as TaCr2O6 in accompanying materials) with a measured modulus near target. Most outputs remain computational.
MatterGen produced a higher fraction of stable unique novel crystals than prior generative baselines in the paper.
outcome · supported
A generated structure was synthesized and measured within 20 percent of a target property.
outcome · supported
primary sources
arXiv · preprint · 6 Dec 2023
Nature · peer reviewed · 15 Jan 2025 · peer reviewed
Secondary domains: Chemistry
13 Sep 2026 · 0.00 → 1.74
Initial adjudicated seed score under methodology 0.1.