Résumé : Inertial sensors are devices capable of measuring the absolute motion of the support they are fixed onto. The advance of very high-end scientific instruments such as gravitational wave detectors, always pursuing ever greater sensitivities and performance, puts a large demand on ultra-high-resolution inertial sensors, capable of measuring very low-frequency and small-amplitude motions. Our group has a long experience in the design of low-frequency inertial sensors intended to be used in active isolation systems. The latest horizontal and vertical interferometric inertial sensors that have been designed have performance that competes with industry standards. They were shown to reach a resolution of 2×10−13 m/Hz−−−√ at 1 Hz and are capable of measuring ground motion from 0.1 to 100 Hz. However, they are large and heavy, measuring approximately 20 × 20 × 30 cm3 each, which makes their integration into practical systems tedious. In addition, experimental characterization of these sensors revealed three main limitations to their resolution. They are: (i) thermal noise, (ii) electronic readout noise and (iii) broadband white noise caused by mechanical and optical nonlinearities. The present paper presents a revised version of the vertical sensor, where the size of the device has been made to fit a 10 × 10 × 10 cm3 while simultaneously addressing the aforementioned sources of noise. The mechanics of the compact sensor is made of a leaf-spring supported pendulum, connected to the frame using a flexure hinge. A moving mirror is connected to the mass and guided using a so-called "4-bar" mechanism, providing the moving mirror with linear translation motion (iii). The joints of the mechanics are made of fused silica, allowing to reach a low natural frequency of ≈1 Hz with a compact design, in addition to significantly reducing structural thermal noise (i) due to the low dissipation rate of fused-silica. On the other hand, the readout system used in this sensor is a homemade design of a Michelson interferometer. The optical scheme features numerous polarizing elements that allow the propagation of two laser beams in phase quadrature, and custom hardware is developed for minimizing electronic noise (ii). Lastly, the vertical sensor is operating in closed-loop, using a homemade actuator design, so as to reduce non-linear effects related to either the mechanics or the optical readout (iii). The sensor frequency response is characterized using a test bench that has been specifically developed for testing low-frequency sensor response. The noise floor is extracted using a Huddle Test.