Quick Question: Has anyone noticed that Haiku 3.5 has 2-3 second additional latency in responses vs sonnet 3.7 when Sonnet shows that it should have 600ms more latency in the preview. It seems like there is something wrong with how VAPI uses Haiku that is adding additional latency making it slower than larger models. Seems like something on the VAPI end. Any solutions to fix this? Thanks in advance