Pinkstack reasoning accuracy model(s) are generally trained/fine tuned to be better at reasoning than original. V1 version, v2 is better overall.