The paper presents MuMAX³, an open-source GPU-accelerated micromagnetic simulation program designed to solve the time- and space-dependent magnetization evolution in nano- to micro-scale magnets using finite-difference discretization. The software is written in Go and CUDA, and is available under the GPLv3 license. MuMAX³ offers high performance and low memory requirements, enabling large-scale simulations on inexpensive hardware. The authors verify each component of the software by comparing results to analytical values and standard problems. MuMAX³ includes extensions such as MFM image generation, moving simulation windows, edge charge removal, and material grains. The paper describes the design, including material regions, geometry, scripting language, and various magnetic interactions. It also discusses the solver constraints, energy minimization, and standard problems. The performance of MuMAX³ is evaluated, showing significant speed-ups compared to CPU-based simulations, and the memory usage is optimized to handle large simulations. The conclusion highlights the advantages of GPU acceleration and the extensibility of MuMAX³.The paper presents MuMAX³, an open-source GPU-accelerated micromagnetic simulation program designed to solve the time- and space-dependent magnetization evolution in nano- to micro-scale magnets using finite-difference discretization. The software is written in Go and CUDA, and is available under the GPLv3 license. MuMAX³ offers high performance and low memory requirements, enabling large-scale simulations on inexpensive hardware. The authors verify each component of the software by comparing results to analytical values and standard problems. MuMAX³ includes extensions such as MFM image generation, moving simulation windows, edge charge removal, and material grains. The paper describes the design, including material regions, geometry, scripting language, and various magnetic interactions. It also discusses the solver constraints, energy minimization, and standard problems. The performance of MuMAX³ is evaluated, showing significant speed-ups compared to CPU-based simulations, and the memory usage is optimized to handle large simulations. The conclusion highlights the advantages of GPU acceleration and the extensibility of MuMAX³.