Exa

The next generation of search

Exa is an applied AI lab building a perfect search engine unlike the world has ever seen.

Exa in numbers

500,000+
developers building on Exa
billions
of pages indexed — news, docs, filings, company profiles, research
1000+
third-party data providers reachable through Exa Connect

Source: exa.ai/enterprise, exa.ai/connect.

The highest quality search API at every latency

Search tier
Accuracy
Latency
Exa Instant
63.6%
323ms
Exa Fast
69.6%
454ms
Exa Auto
67.8%
~1.2s
Exa Deep
80.0%
4.3s

Latency is a dial, not a constant — the same API serves a live lookup and an overnight enrichment run. And it returns excerpts rather than whole pages: Highlights reports ~94% fewer tokens on some evals.

Accuracy on the MultiLoko dataset; source: exa.ai/products/search. Token figure: exa.ai/blog/highlights-for-agents. Exa's enterprise page reports different figures for the same tiers — those are measured on SealQA Hard, not MultiLoko.

Exa case studies

"Traditional search engine solutions weren't enough for us. Exa powers all parts of Devin."

Walden Yan — Co-founder, Cognition

"Exa is necessary for agents on HubSpot to supplement internal data with high quality, real-time people and company search."

Praty Sharma — AI Lead, HubSpot

"Standard signals like funding, job changes, and hiring are commoditized. Exa is how we find leads surfacing the triggers that actually predict a good prospect."

Prabhav Jain — CEO, 11x

"We're entering a new era where software actually does the work. For that to be real, the agents need live intelligence behind them. Exa is where that comes from."

Daniel Lereya — Chief Product and Technology Officer, monday.com

All quotes verbatim from exa.ai/customers.

Exa products

Search API
Accurate real-time web search
Contents API
Token-efficient page extracts
Agent API
Frontier web research in one API
Monitors API
Track changes on the web
Exa Connect
Access third-party data providers
Exa Vertical

Demo

Search: accuracy without paying for it in latency

On the hardest retrieval benchmark, the alternatives make you pick. Exa is 17× faster than Parallel at higher accuracy, and more than twice as accurate as the providers in its own latency class.

Search quality on SealQA Hard: accuracy versus P50 latency. Exa 45% at 178ms, Parallel 37% at 3051ms, Brave 20% at 559ms, Tavily Ultra-Fast 11% at 260ms.

Source: exa.ai/blog/exa-instant, 12 Feb 2026. Dataset: SealQA Hard. Exa's own benchmark.

Contents: the same answer, a fraction of the context

Highlights reaches 90% accuracy on 500 characters. Truncated full text needs 8,000 to match it. That is the same answer for 16× less context — the token cost of every downstream LLM call.

SimpleQA accuracy versus characters. Highlights: 90% at 500 chars rising to 93% at 8k. Truncated full text: 52% at 500 chars rising to 90% at 8k.

Source: exa.ai/blog/highlights-for-agents, 22 Apr 2026. Dataset: SimpleQA.

Thank you