Can't be less though, the latency difference between them implies that GPT-4 is significantly larger. This is unlike with PaLM, where latency decreased.