Tested the Latency of Various AI Search Tools — Interesting Results

Today, I conducted a quick side-by-side comparison to see how quickly different AI search tools respond. Here are the results (P50 & P95 times, US West):

• Exa: ~423 ms (P50), ~604 ms (P95) - the fastest in the test, it felt almost instantaneous.

• Brave Search API: ~717 ms (P50), ~1,098 ms (P95) - reasonably fast, but you can notice a slight delay.

• Google Programmable Search: ~1,044 ms (P50), ~1,454 ms (P95) - noticeably slower, particularly at higher percentiles.

I wasn't expecting such a significant gap between the fastest and slowest tools, especially since they all perform a similar function of finding relevant sources. This experience highlighted how much sub-second latency can influence the usability of an AI tool, especially when chaining multiple searches together in agents or workflows.