Physics-based simulators for neuromechanical control of virtual animals have the potential to significantly enhance our understanding of intricate structure-function relationships in neuromuscular systems, their neural activity and motor control. However, a key challenge is the accurate prediction of the forces that muscle fibers produce based on their complex patterns of electrical activity ("spike trains") while preserving model simplicity for broader applicability. In this study, we present a chemomechanical, three-dimensional finite-element muscle model - JiSuJi (pronounced , meaning "ultrafast muscle" in Chinese) - that efficiently and accurately predicts muscle forces from naturalistic spike trains. The model's performance is validated against songbird vocal muscles, a particularly fast and therefore challenging muscle type. Our results demonstrate that JiSuJi accurately predicts both isometric and non-isometric muscle forces across a variety of naturalistic neural activity patterns. JiSuJi furthermore outperforms state-of-the-art muscle simulators for accuracy, while maintaining computational efficiency. Simulating muscle behavior offers a promising approach for investigating the underlying mechanisms of neuro-muscular interactions and precise motor control, especially in the fast-contracting muscles of animal model systems.