Introducing Exa Instant
Introducing Exa Instant
Today we're introducing Exa Instant - the fastest web search engine in the world. Exa Instant is sub-200ms – that's faster than Google search. This is particularly valuable for realtime AI chat and voice applications.
Evals
We compared Exa with many providers. Exa Instant was faster by up to 15x.
We benchmarked all providers from a datacenter in us-west-1 (northern california). The network latency for Exa, for example, was roughly 50ms. For the query dataset, we used SealQA queries (and concatenated some random words generated by GPT-5 to avoid any caching).
We also measured quality evals. Exa Instant maintains quality standards. Of course, for the highest quality search, higher compute options like Exa Fast, Exa Auto, and Exa Deep are available. The providers tested were Tavily Ultra Fast, Brave and Parallel one-shot.
Who needs Instant?
Humans don't notice search that's faster than half a second. But for AI agents, every millisecond matters.
That's because AI agents use search as part of a workflow. If the whole workflow needs to be under a second, then the search part needs to be near instantaneous in order not to be a bottleneck.
Exa powers thousands of companies, and Exa Instant should help many of them. Particularly:
Deep research or coding agents: If a deep research agent makes 50 search calls and each one is 200ms faster, that's 10 seconds of savings for users.
Low-latency products: Chat apps and AI voice companions are very latency sensitive. As LLMs get faster, the search needs to be in proportion.
How we built it
Building the fastest search in the world means searching over tens of billions of pages in under 200ms. That requires optimizing every part of our search and retrieval stack.
Most search APIs are far slower, mostly because they need to wrap Google SERP under the hood. That means they send queries to server farms that use Google to return the results. This takes over 700ms P50, and so any search API that wraps Google has a minimum 700ms P50.
Where search is going
Most people don't realize – even in San Francisco – that search as an industry is going through a massive overhaul.
AIs will soon search the web more than humans, and these AIs have a whole set of different needs. Low latency is just one of them.
We have more launches coming soon that address these needs. Stay tuned :)
How to use Exa Instant
Test it out at dashboard.exa.ai by selecting Search Type → Exa Instant. Docs here.
If you think 200ms is not instantaneous enough, we agree. Come help us make it faster. We're hiring.
