Small Language Models: Same Performance But Cheaper