MatterSim is a deep learning model designed to predict materials properties across a wide range of elements, temperatures, and pressures. It leverages large-scale first-principles computations to learn and simulate atomistic behaviors, achieving up to ten times the accuracy of previous methods. The model serves as a machine learning force field, capable of predicting ground-state structures, energetics, and dynamics under realistic conditions. MatterSim covers a broad compositional and configurational space, enabling it to make accurate predictions for a wide range of applications, including materials discovery, phonon prediction, mechanical property prediction, and phase diagram construction. It also demonstrates high data efficiency, requiring minimal additional training data to fine-tune for specific systems. The model's performance is benchmarked against first-principles methods, showing significant improvements in accuracy and generalizability. MatterSim's ability to predict free energies with sub-10 meV/atom error and its capability to construct temperature- and pressure-dependent phase diagrams open new avenues for materials design and discovery.MatterSim is a deep learning model designed to predict materials properties across a wide range of elements, temperatures, and pressures. It leverages large-scale first-principles computations to learn and simulate atomistic behaviors, achieving up to ten times the accuracy of previous methods. The model serves as a machine learning force field, capable of predicting ground-state structures, energetics, and dynamics under realistic conditions. MatterSim covers a broad compositional and configurational space, enabling it to make accurate predictions for a wide range of applications, including materials discovery, phonon prediction, mechanical property prediction, and phase diagram construction. It also demonstrates high data efficiency, requiring minimal additional training data to fine-tune for specific systems. The model's performance is benchmarked against first-principles methods, showing significant improvements in accuracy and generalizability. MatterSim's ability to predict free energies with sub-10 meV/atom error and its capability to construct temperature- and pressure-dependent phase diagrams open new avenues for materials design and discovery.