Anti-spring mechanisms are widely used for improving the noise performance of MEMS accelerometers due to their stiffness softening effect. However, the existing mechanisms typically require large bias force and displacement for achiev¬ing stiffness softening, leading to large device dimensions. Here, we propose a novel anti-spring mechanism composed of two pre-shaped curved beams connected in a parallel configuration, which can achieve stiffness softening without requiring large bias force and displacement. The stiffness softening effect of the mechanism is verified through theoretical modeling and finite element method (FEM) simulation. After that, the mechanism is implemented in a 4.2 mm × 4.9 mm MEMS capacitive accelerometer prototype. The experimental results reveal that the sensitivity of the accelerometer increases by 10.4% compared to the initial sensitivity, at the same time, the noise floor and bias instability decrease by 10.5% and 4.2%. The sensitivity,nonlinearity, bias instability, and noise floor after biasing are 51.1 mV/g, 0.99%, 0.24 mg, and 21.3 μg/√Hz, respectively. Thus, the proposed mechanism can enhance the performance of the accelerometer. This work provides an innovative approach for improving the performance of MEMS accelerometers while enabling miniaturization.