Without knowing more, one possibility is that the integration of the MPU6050 data includes a constant (possibly an offset from zero) that is consistently being integrated, while the Simulink data does not.
I obviously have no way of checking that, given only the posted information.
EDIT — (16 Sep 2023 at 13:29)
One option to correct it might be to subtract the first element of the vector from the entire vector, then integrate that. (Create a copy of the original vector to do that, so that you retain the original vector.)
