It IS being smoothed, it has a different base smoothing factor than on simulated client side, which is settable in the character movement component itself.
Listen server has: ListenServerNetworkSimulatedSmoothLocationTime and ListenServerNetworkSimulatedSmoothRotationTime
The difference in smoothing (assuming you are looking at meshes actually attached to the default mesh of a character because otherwise they aren’t smoothed) is because the servers smoothing time is 0.04 seconds while the clients is 0.1 seconds. When movement ends the transition to capsule location is significantly quicker on the listen server side than the clients side, but this is changeable with those two variables above.
The REAL difference between server / simulated client side is that the simulated client is predicting velocity which means that their smoothing tends to predict “forward” of momentum while the server is lerping between states so its smoothing predicts “backward” from momentum. This is visible when looking at the character stopping as the mesh lerps backward from the direction of velocity into its correct resting pose on the client while on the server it lerps forward towards the direction of velocity into its resting pose.
Not entire sure if the above smoothing difference is intended? Its more jarring on client side to have the character stop and bounce backwards from its momentum but it retains its position closer than the server does during the actual movement itself.