Exa Highlights Upgrade: 500-Character Summary Matches 8000-Character Full Text Accuracy, Token Usage Reduced by 16 Times

According to monitoring by Dongcha Beating, AI search API company Exa has announced a significant upgrade to its Highlights feature. Highlights is Exa’s capability to extract dense summary snippets relevant to queries from web pages, generated in real-time (not cached) within 100ms for each request. In the SimpleQA benchmark test, Exa Highlights with 500 characters matched the accuracy of the original 8000-character full text, with token consumption reduced by 16 times; 4k character highlights performed better than 32k character full texts. The improvement is particularly notable in long document scenarios such as technical coding documentation, API references, SDK documentation, and research papers: with a 500-character budget, highlights achieved an accuracy of 60%, while the full text only reached 6%. Exa’s own agent endpoints, such as /answer, Deep, and Websets, utilize Highlights as the underlying retrieval mechanism without reading the original page content. Highlights is not a new feature but an existing content extraction option in the Exa API. This upgrade is at the model level, and all requests that enable the enableHighlights parameter will automatically benefit without requiring code changes.