tech
January 28, 2026
Exa AI Research Blog
Discover the latest in AI research and semantic search technology on the Exa blog. Learn how our neural network search engine provides high-quality web data for AI applications
TL;DR
- A new benchmark has been created for company search, focusing on retrieval of structured data over memorized knowledge.
- The benchmark includes an ~800-query dataset and an open-sourced evaluation harness.
- It distinguishes between static facts (e.g., founding year) and dynamic facts (e.g., employee count, funding).
- Two evaluation tracks are included: Retrieval (testing company retrieval for queries) and RAG (testing fact extraction from retrieved content).
- The dataset was designed to avoid well-known large companies and instead focus on regional players, smaller companies, and niche verticals to ensure retrieval is necessary.
- The benchmark is part of a larger effort to build an evaluation ecosystem for various entity types and search domains.