This is a quick that is being resolved as we speak. We have implemented COSMO analytical gradients, but not the Hessians so far. And, we are not properly catching this in the code. If a frequency calculation with COSMO is called the code should use analytical gradients to form the numerical hessian.