On that note it would be nice to get these benchmark numbers based on the different reasoning settings.