Exa 3.0 search API outperforms competitors on 8 benchmarks

Exa has launched Exa 3.0, an upgraded web search API for AI agents, featuring a larger embedding model and a cost-effective search type called Exa Lite. The API also includes Dynamic Highlights, now generally available, which uses 95% fewer tokens than full page text while improving answer quality. Exa claims to lead Perplexity, Brave, and Parallel on 8 benchmarks, with results from its own testing. The benchmarks were run by Exa, making them self-reported. The new search types recommended by Exa are Lite, Auto, and Deep, with legacy types still functional. Exa compared its highest-scoring endpoint, Exa 3.0 Auto, against Perplexity Web, Brave LLM Context, and Parallel Advanced, all evaluated in October 2026. Exa's results showed significant leads in ATLAS and WebCode benchmarks, with margins of 4.8 and 5.6 points over Perplexity, respectively. Other benchmarks showed smaller advantages. Exa also highlighted cost efficiency, with Lite scoring 79 at $0.045 per task and Auto scoring 80.5 at $0.065 per task, setting a new Pareto frontier. Exa's data includes 1.27B ecommerce listings, 211M+ local businesses, and real-time news, sports, and weather data. The Exa Library organizes this into searchable collections, including 320M people pages and 500M research paper pages.