3 ms·

By removing "Answer only with the result" all models can answer this correctly by doing "chain of thoughts"