The Least Squares Moving Particle Semi-Implicit (LSMPS) method theoretically achieves high-order accuracy, greatly enhancing simulation precision and stability. However, the method’s complex numerical discretization and high computational cost limit its applicability in simulations involving a large number of particles and complex scenarios. To address this, a GPU-accelerated weakly compressible LSMPS method with phase transition was proposed. By adhering to the conservation laws of mass, momentum, and energy, the gas volume generated during particle evaporation is dynamically separated into independent gas particles, with the separation direction aligned to the interface normal pointing toward the gas phase. The volume change due to evaporation is incorporated as a source term in the Pressure Poisson Equation (PPE). OpenACC parallel directives were added to guide the parallelization and memory management, GPU acceleration was implemented within each time step. Benchmark simulations of two-dimensional horizontal film boiling validated the accuracy and stability of the proposed method in interface tracking and heat and mass transfer during phase transitions. In the two-dimensional horizontal film boiling simulation involving 30,000 fluid particles, the bubble morphology and heat transfer results were consistent with reference. GPU acceleration achieved over a 30-fold speedup compared to the serial simulation. This method is particularly effective for simulating multiphase flows with dynamic phase interfaces, offering a simple, reliable, and scalable solution for simulations involving a large number of particles.