OPEN LINK ↗
// 1h agoPRODUCT LAUNCH
Exa 3.0 makes agent search cheaper, richer.
Exa 3.0 upgrades Exa’s search API with a larger embedding model, billions of new documents, Dynamic Highlights, and a $4-per-1,000-request Lite tier ([official announcement](https://exa.ai/blog/exa-3)). Dynamic Highlights reportedly cuts token usage by 95%, while Exa claims top results across eight benchmarks.
// ANALYSIS
The biggest win is economics: better context selection can reduce downstream model spend more than marginal retrieval gains. The benchmark lead is promising but self-reported, so developers should measure cost per correct task on their own workloads.
- –Dynamic Highlights selects tokens across results, reducing repetition and potentially improving grounding.
- –Lite and Auto create a clear cost-quality ladder at $4 and $7 per 1,000 requests.
- –The expanded index adds ecommerce, local businesses, news, weather, sports, flights, and other structured data.
- –Early benchmarking suggests Dynamic Highlights may increase latency even as it reduces token usage ([AI Primer](https://www.ai-primer.com/engineer/stories/exa-3-search)).
- –Exa’s benchmark comparisons are vendor-run, so teams should validate performance against their own queries.
// TAGS
exa-3-0searchapiembeddingagentragpricing
DISCOVERED
1h ago
2026-10-20
PUBLISHED
7h ago
2026-10-20
RELEVANCE
9/ 10