Tin-based halide perovskites have emerged as promising lead-free alternatives for optoelectronic applications, yet their structural stability and phase behavior at finite temperatures remain challenging to predict. Here, we assess the predictive capabilities of the foundational machine learning model MACE-MP-0 – trained on a broad chemical space and applied without system-specific fine-tuning – for the temperature-dependent behavior of \(\hbox {CsSnBr}_{3}\) and \(\textrm{Cs}_{2}\textrm{SnBr}_{6}\) . Molecular Dynamics simulations in the NpT ensemble were performed from 100 K to 500 K, and thermodynamic and structural descriptors including enthalpy, specific heat, radial distribution functions, translational order, bond angle distributions, and vibrational spectra were analyzed. Our results show that \(\hbox {CsSnBr}_{3}\) undergoes a sequence of low-temperature structural transitions from an orthorhombic to an intermediate tetragonal phase, followed by a transition to the cubic phase, as evidenced by the evolution of lattice parameters and anomalies in enthalpy and specific heat. In contrast, \(\textrm{Cs}_{2}\textrm{SnBr}_{6}\) remains cubic and maintains a more rigid octahedral framework across the entire temperature range. Overall, MACE-MP-0 qualitatively reproduces the key thermal and structural features of these materials without fine-tuning, while remaining discrepancies in the transition temperatures when compared to experimental findings indicate that system-specific refinement with Density Functional Theory data could further improve quantitative agreement.
