From our research, it seems like there’s no “best” one, only the most suitable for your use case.

We included a scorecard here that might help you to balance the trade off: https://telem.ai/blog/latency-research#sec-scorecard