MIT-Developed Nequix MP Model Achieves Top-Three Performance with 75% Lower Training Costs for Materials Simulation

Cambridge, MA – Teddy Koker, a PhD student at MIT, has announced the introduction of Nequix MP, a new foundation model designed for materials science. The model aims to significantly enhance atomistic simulations, offering a more efficient and accessible approach to understanding material behavior. Koker highlighted its capabilities, stating on social media, > "Introducing Nequix MP, a new foundation model for materials. Example of using Nequix to simulate the dissolution of a NaCl nanocrystal in water."

Nequix, a compact E(3)-equivariant potential, leverages a simplified NequIP design combined with modern training practices, including equivariant root-mean-square layer normalization and the Muon optimizer. Developed in JAX, this innovative architecture focuses on retaining high accuracy while substantially reducing computational requirements. The model has 700,000 parameters, demonstrating a lean yet powerful design.

The model's performance has been rigorously evaluated on key benchmarks, where it has shown impressive results. Nequix ranks third overall on both the Matbench-Discovery and MDR Phonon benchmarks, positioning it among the top-tier solutions in the field. Crucially, Nequix was trained in just 500 A100-GPU hours, requiring less than one-quarter of the training cost of most other methods. Furthermore, it delivers an order-of-magnitude faster inference speed compared to the current top-ranked model.

This cost-effective and high-performance approach is poised to democratize access to advanced atomistic modeling. By providing a practical alternative to more computationally intensive large-scale foundation models, Nequix MP could enable a broader range of researchers and institutions to conduct high-quality materials simulations. Future development is expected to explore further scaling, pretraining, fine-tuning regimes, and additional cost reduction strategies, promising continued advancements in the field.