Under the same conditions I understand that gpt-3.5-turbo is faster (returns outputs with lower latency).