Modes

Every search endpoint takes a mode. It sets how many candidates are judged and whether the best ones are opened and read before they are ranked.

normal is the default. Headlines already judged for your query don’t count against the limits above, and aren’t paid again.

ultra is the cheapest mode. Against Exa’s list prices, checked September 25, 2026: ultra costs 7× less than an Exa search ($7, which includes the text of up to 10 results) and fast 5× less, normal 3.1× less than that same search, and deep 2.1× less than an Exa deep search ($12–15). For headline search alone, Perplexity’s Fast Search ($1 per 1,000 requests of up to 5 queries) and Parallel’s fast search ($1) cost the same as ultra and less than fast. Every comparison, mode by mode and with its source, is on the pricing page.

Choosing a mode

• ultra judges the headline alone and is the cheapest. Use it for high-volume classification, alerts, and anything where the headline says it all.
• fast is the best default for agents in a loop: quick, cheap, and good recall when the headline names the topic.
• normal reads the top results before ranking them. Use it when a wrong result is expensive — a headline that promises more than its article is caught and moved to rejected.
• deep widens the topic with synonyms and related terms, judges more headlines and reads more articles, with highlights and essentials included. Use it for research, reports and questions where you need everything.

When the index falls short

If the index returns fewer than 3 good results, typesearch looks for sources beyond it, within a time budget: at most 0.8 s more in ultra and fast (the rest continues in the background, so the next search benefits), 3 s in normal and 8 s in deep, which can also use the search box of the sites it proposes.

Sources are proposed by a model that understands the topic of your query and by an open index of worldwide headlines built with data from the GDELT Project. Articles are always read on the original site, following its robots.txt, and judged by the same relevance model. No third-party search engine is involved.

Results found this way come with found_in: "discovery", and the response says what happened in discovery (used, background, skipped, budget or timeout). If time ran out, incomplete is true: repeating the search a little later usually brings more.

Results arrive in two stages

In every mode, results judged from the headline arrive first — about a second in — and each one is confirmed in place as its article is read. With streaming you get those partial results as partial events; without it, you get the final result.

What doesn’t cost twice

• Judgment memory. What the model already said about a headline for a query is reused for a week, so a repeated or similar search only pays for what is new.
• Result cache. An identical request within 10 minutes is served from cache: usage.tokens is 0 and it is not billed.

See Costs and caching for hard caps per request.