Name one use case where there is a difference between latency of 200 t/s (fireworks.ai mixtral model) and 500 t/s (groq mixtral)? Not throughput and not time to first token, but latency.

Groq model shines at latency, not at the other two.