Overview
Exa is a search API built for AI agents, dedicated to providing agents with real-time web data retrieval capabilities. Its search categories cover the entire web, news, companies, research, people, and finance, among other vertical domains, and its output supports three formats—Results, Answer, and Structured—making it flexible to call in different scenarios. On October 19, 2026, Exa released version 3.0. The Search API added the Dynamic Highlights mode, which can decide on its own how much text to return per page within a shared budget; the company says this can reduce costs by about 21%. The product is offered via API and dashboard and includes playground debugging features, making it easy for developers to quickly integrate and validate.
Key Features
- Real-time web data retrieval: Provides AI agents with real-time web data search capabilities, supporting vertical categories such as the entire web, news, companies, research, people, and finance to meet diverse information acquisition needs.
- Dynamic Highlights mode: Newly added to the Exa 3.0 Search API, this mode can decide on its own how much text to return per page within a shared budget; the company says this can reduce costs by about 21%.
- Multiple latency tiers: Offers three latency tiers: Lite at about 450ms, Auto at about 1s, and Deep at about 10s, allowing users to flexibly choose based on the speed and depth requirements of the scenario.
- Three output formats: Supports three output formats—Results, Answer, and Structured. Examples show that compared with raw web pages, it can reduce tokens by about 61%, helping to control context overhead.
- Vertical domains and batch lookup: Covers vertical domains such as code, finance, and GTM intelligence. The FindAll interface can batch-find matching companies and people by criteria, improving the efficiency of structured information acquisition.
- Benchmark performance and debugging support: The company says Exa 3.0 leads on three search agent benchmarks—BrowseComp, DSQA, and WideSearch—with a comparison chart showing 91.8% accuracy; it is tied for first place on the Artificial Analysis Search Index. The product includes playground debugging for quick validation.
Use Cases
- When AI agents need real-time web information, they use the Exa API to obtain the latest web page data and news content.
- When building research-oriented or question-answering agents, developers use Answer or Structured output to reduce token consumption and improve response efficiency.
- Finance and GTM teams use vertical categories and the FindAll interface to batch-retrieve matching company and people information.
- Latency-sensitive applications can choose the Lite tier, while scenarios requiring deep retrieval can choose the Deep tier.
- Quickly debug search requests through the dashboard and playground to validate the effects of different categories and output formats.
Pros
- Designed specifically for AI agents, providing a real-time web data retrieval API with a clear integration path.
- Search categories cover multiple vertical domains such as the entire web, news, companies, research, people, and finance.
- Exa 3.0 adds the Dynamic Highlights mode, which the company says can reduce costs by about 21%.
- Offers three latency tiers—Lite, Auto, and Deep—making it easy to balance speed and depth by scenario.
- Output supports three formats—Results, Answer, and Structured—and examples show it can reduce tokens by about 61%.
- The company says it leads on the BrowseComp, DSQA, and WideSearch benchmarks and is tied for first place on the Artificial Analysis Search Index.
Pricing
Billed per request. Search endpoints cost $4 (Lite), $7 (Auto) and $12 (Deep) per 1,000 requests with up to 10 results, Answer $5, Monitors $15, and Contents $1 per 1,000 pages; each additional result beyond 10 costs $1 per 1,000 requests, and AI page summaries cost $1 per 1,000 pages. Agent runs are priced from $0.012 to $1.00 by effort level — Minimal $0.012, Low $0.025, Medium $0.10, High $0.50, X-high $1.00, Ultra usage-based — with usage components billed separately: $0.10 per Agent Compute Unit, $0.005 per search call, $0.02 per email enrichment and $0.07 per phone enrichment. The free tier includes $10 in monthly credits plus a $10 onboarding bonus, 10 QPS and 50 agent concurrency; pay-as-you-go raises this to 25 QPS; enterprise adds custom concurrency, zero data retention and SSO. Refer to the official site for current rates.
Summary
Exa is a search API for AI agents, providing real-time web data retrieval and support for multiple vertical categories. Exa 3.0 introduces the Dynamic Highlights mode, which the company says can reduce costs by about 21%, and offers three latency tiers and three output formats. The company says it leads on multiple search agent benchmarks and is tied for first place on the Artificial Analysis Search Index. The product is offered via API and dashboard and includes playground debugging, making it suitable for integration into agent applications that need real-time web data.
Version History
- Exa 3.0 released: Search API adds Dynamic Highlights with about 21% lower cost (2026-10-19): Exa released version 3.0 on 2026-10-19, adding a Dynamic Highlights mode to its Search API that decides how much text each page returns within a shared budget, which the company says cuts cost by about 21%. Exa says 3.0 leads on BrowseComp, DSQA and WideSearch, the standard benchmarks for search agents, and ties for first on the Artificial Analysis Search Index. Search latency comes in three tiers: Lite at about 450ms, Auto at about 1s and Deep at about 10s.