5.2 seems worse on overfitting for esoteric logic puzzles in my testing. Tests using precise language where attention has to be paid to use the correct definition among many for a given word. It charges ahead with wrong definitions in a far lower accuracy and worse way now.