All models that allow a reasoning or thinking mode should be activated, ideally through a unified interface.