Does anyone have an idea how retell manages to have low latency for gpt 5.4 mini? I tested both for vapi and retell and on retell it was like 600ms+ faster. But any other model was similar latency.