This is a remarkable result if confirmed independently. The gap between solving competition problems and open research problems has always been significant - bridging that gap suggests something qualitatively different in the model capabilities.