Exa Highlights

This post details improvements to Exa's highlights system, which extracts relevant excerpts from web pages for agentic search. The new highlights offer higher quality results for ~94% fewer tokens on some evals, significantly reducing costs and leading to latency benefits. On benchmarks like SimpleQA, 500 characters of Exa's highlights match the accuracy of the first 8000 characters of the page, and use 16x fewer tokens. The highlights model is an in-house method that runs for every request (not cached!) and completes in under 100ms. Improvements have been made on technical coding docs and long-context documents like API references, SDK docs, specs, and research papers. These highlights are used as the retrieval substrate in Exa's agentic endpoints (/answer, Deep, and Websets).