14 Oct 2026
What an AI task costs: a 200-fold spread across 38 priced models on our board
A 50,000-token summary costs a quarter of a cent on the cheapest text model on our board and 60 cents on the dearest. We priced five tasks…
Exa's web search API for AI agents, rebuilt on a larger embedding model with a bigger index, shorter page highlights and a new Lite search type at $4 per 1,000 requests.
Not yet independently verified. The benchmark results are Exa's own and the post does not say who ran them; prices and index figures come from Exa's announcement of 19 October 2026 and were not independently checked. We will update this when it can be confirmed, and remove this note.
Exa 3.0 is the third generation of the web search API that Exa Labs sells to developers building AI agents and applications that need to look things up on the internet. Exa announced it on 19 October 2026 and made it available through its API the same day.
Exa attributes the release to three changes in how it retrieves pages. It trained a larger embedding model, the part that matches a query to pages by meaning, with more compute and data than any of its earlier models. It made Dynamic Highlights generally available: instead of returning a whole page, the API returns the passages that answer the query, which Exa says uses 95% fewer tokens than full page text, a saving that matters when every token passed to a language model is paid for. And it added billions of documents across structured kinds of data, including products, local businesses, flights, events, sports and weather.
By Exa's count the index tracks more than 2 trillion URLs and serves 100 billion pages, with 1.27 billion ecommerce listings carrying prices and availability and more than 211 million local businesses with hours and ratings, alongside real-time news, sports, weather and market data. Exa publishes a chart of eight search benchmarks on which it places Exa 3.0 ahead of Perplexity, Brave and Parallel; it gives a BrowseComp score of 91.8%, but the post does not say who ran the tests, so these are the maker's own results.
The API is built for software rather than people: agent frameworks, research assistants and products that need fresh facts from the web inside a workflow. The new Lite search type is the cheaper option for high-volume calls, and Auto is the default.
Maker: Exa Labs Inc. Announced: 19 October 2026 Pricing: Exa Lite $4 per 1,000 requests; Exa Auto $7 per 1,000 requests; Dynamic Highlights and the expanded data sources at no extra charge Access: API
No questions yet. Be the first to ask about Exa 3.0.
Everything here links back to the same verified catalogue. Pick your next stop.