The calculation results of different versions (v1.3.2 vs v1.4 vs v1.4.1) of Psi4 vary greatly

I used Psi4 to calculate the energy of molecules in Conformer Benchmark dataset ( https://github.com/ghutchis/conformer-benchmark/ 702 molecules, each molecule has about 10 conformations).

I tried three versions of Psi4 respectively, including V1.3.2, V1.4 and V1.4.1, and then calculated the correlation between the results of the three versions. The results are shown in the following table. The consistency is not high, especially v1.4.1, which is greatly different from the other two versions. Why are the calculation results of different versions so different?

In addition, I also calculated the correlation between Psi4 results and Orca results (from Conformer benchmark), the correlation is not high, why?

my code:

def run_psi4_calc(xyz_str, dtype='xyz', basis_set='WB97X-D/def2-SVP', reference='uks', max_iter=100):

    psi4.set_options({
      'reference': reference,
      'maxiter': max_iter
    })

    mol = psi4.core.Molecule.from_string(xyz_str, dtype=dtype)
    energy = psi4.energy(basis_set, molecule=mol)

    return energy



# get molecular 3d coordinates from Conformer benchmark
xyz_str = get_xyz_str()

energy = run_psi4_calc(
    xyz_str=xyz_str,
    dtype='xyz+',
    basis_set='WB97X-D/def2-SVP',
    reference='uks',
    max_iter=100
)

Closed. Please don’t create a forum topic and a GitHub issue.