AI API relay vs official API: latency, reliability and cost comprehensive analysis