# Best web search API: ranked by live price The best web search API depends on what your agent reads next. Links and snippets suit an agent that opens pages itself, while excerpts or page text in the answer save a second call. This page ranks the looot catalog providers by live price per search, so the cheapest one that returns what you need comes first. ## Live prices Prices read on 2026-10-28 from the live looot catalog. They change when a provider changes its price. Cheapest today for web results for a query: Limadata at $0.0020 per call. | Rank | Provider | Price per call | Per 1,000 calls | |---|---|---|---| | 1 | Limadata (limadata) | $0.0020 per call | $2.00 | | 2 | Jina AI (jina) | $0.0025 per call | $2.50 | | 3 | Keenable (keenable) | $0.0040 per call | $4.00 | | 4 | Brave Search (brave-search) | $0.0050 per call | $5.00 | | 5 | Linkup (linkup) | $0.0050 per call | $5.00 | | 6 | Parallel (parallel) | $0.0050 per call | $5.00 | | 7 | You.com (youcom) | $0.0050 per call | $5.00 | | 8 | Octen (octen) | $0.0060 per call | $6.00 | | 9 | Exa (exa) | $0.0070 per call | $7.00 | | 10 | Tavily Search and Extract API (tavily) | $0.0080 per call | $8.00 | | 11 | Olostep (olostep) | $0.0090 per call | $9.00 | ## Providers - limadata: Returns Google results for a query, page by page. It takes a query and a page number and nothing else, so there are no country or date filters. Also: Page to markdown, Company enrich. - jina: Searches the web and returns the top results read into text, so an agent gets page content in one call. Answers are slower than a links-only search. Also: Page to markdown, Rerank. - keenable: Searches with a deep or a realtime mode, and filters by site and by the date a page was published or fetched. It returns results with snippets whose length you set. Also: Page to markdown. - brave-search: Searches Brave's own web index and returns titles, links and descriptions. Page text is not in the answer, so an agent that needs it makes a second call to a scraper. Also: News, Images, Videos. - linkup: Takes a question in plain words and returns results with page content, a sourced answer or JSON in a schema you send. You choose fast, standard or deep, and deep costs more per call. Also: Answer, Research. - parallel: Takes one to three short search queries plus an objective in plain words, and returns links with excerpts chosen for that objective. A single plain query string is not its input shape. Also: Extract, Research task. - youcom: Returns web and news results in separate sections, with country, language, freshness and domain filters. Its research endpoint answers longer questions. Also: Research, Page to markdown. - octen: Returns titles, links, highlights and publish dates, with optional full page content and filters for time, language and domains. Octen has no other capability in the catalog. - exa: Runs a neural search that matches the meaning of the query, not only its words, and returns links with optional page text. A plain keyword query can rank differently than it would on Google. Also: Answer, Similar pages, Article text. - tavily: Returns results built for an agent to read, each with a content snippet, and can add a short answer. It has extract, crawl and site map endpoints for the next step. Also: Extract, Crawl, Site map. - olostep: Returns a deduplicated list of links for a query, with a fast mode that runs the query word for word and domain filters. Page text comes from its scrape endpoints. Also: Answer, Page to markdown. ## How we rank Every provider here has a web search endpoint in the looot catalog, billed per call. We rank by the price per call of that endpoint, read live from the catalog. A cheaper search ranks higher. Search endpoints billed per result are left out, because their price per call depends on how many results you ask for, so the units do not compare. Some providers price by mode or depth. The ranking reads the price the catalog lists for the endpoint, and a deeper mode can cost more. The catalog publishes measured run counts in its search results, but this page does not rank by them. Where a provider has few runs, one bad answer moves the figure a lot. Extra capabilities do not make a provider rank higher. Providers outside the looot catalog are not ranked. ## Cheapest web search API per 1,000 searches On 2026-10-28 the cheapest web search API in the looot catalog is Limadata at $2.00 per 1,000 searches ($0.0020 per call), then Jina AI at $2.50 and Keenable at $4.00. The dearest, Olostep, costs $9.00 per 1,000 searches. The ranking shows each provider's live price per search and per 1,000 searches. Read it before a batch, because a mode or depth setting can move one provider to a dearer price. A cheap links-only search plus a page read can cost more than one search that returns page text. Add both prices when your agent reads every result. Web search API for agents, step by step: /use-cases/web-search-api ## Web search API for AI agents An agent needs short, clean text it can quote, not a page of ads and widgets. Providers such as Tavily, Exa, Linkup and Parallel return snippets or excerpts written for a model to read, and Jina returns the page text itself. If your agent already has a scraper, a links-only search such as Brave or Limadata is enough. It picks the links, and the scraper reads them. ## Choosing one - Links and descriptions from an independent index: brave-search - Search by meaning, then find similar pages: exa - A sourced answer or JSON in your own schema: linkup - Excerpts focused on a goal you describe: parallel - Page text for every result in one call: jina - Google results with no extra settings: limadata ## Prompt Use looot to search the web for the five newest articles on email verification APIs: call the cheapest web search endpoint, list title, link and one sentence on each result, then tell me what the search cost. 1. Set up https://looot.ai/skill.md. 2. Search the catalog for each step and check its price before you call. 3. Run the cheapest endpoint that fits. 4. Show the results as a table and the total cost. ## What is the best web search API? It depends on what your agent reads next. This page ranks the looot catalog providers by live price per search. Pick the cheapest for links and snippets, or a provider that returns page text or an answer when you would otherwise pay for a second call. ## What is the cheapest web search API? The first row of the ranking above, read live from the catalog. Prices can change, so the page reads them on each refresh and shows no number when the catalog is unreachable. ## Is there a free web search API? This page ranks priced endpoints. Some catalog search endpoints bill per result and are left out here, because their price per call depends on the number of results. ## What is the best web search API for AI agents? One that returns text a model can read without a second call. Tavily, Exa, Linkup and Parallel return snippets or excerpts, and Jina returns page text. A links-only search suits an agent that already has a scraper. ## What are the best Google Search API alternatives? Google's own web search API is not in the looot catalog, so this page does not rank it. Limadata returns Google results, the providers above search their own or other indexes, and the best SERP API page ranks Google results with positions. ## What is the difference between a web search API and a SERP API? A SERP API returns a search engine's results page as data, with positions, ads and features, which suits rank tracking. A web search API returns results for an agent to read, often with snippets or page text, and the order is the provider's own. ## What does a web search API return? A list of results for a query, each with a title and a link. Most add a snippet, some add excerpts, page text, publish dates or a short answer with sources. ## How do I get the text of each result? Pick a provider that returns page text, such as Jina or Octen with full content on, or send each link to a page to markdown endpoint. Most providers here have one, shown as a tag in the ranking. ## How do I check the same searches every week? Run the same queries again and compare the links. Doing it on a timer needs scheduled runs, not available yet, so run it by hand when you want fresh results. ## How does an agent call a web search API? Paste the prompt on this page into your agent. It sets up looot, finds the web search endpoints, checks the price and runs the cheapest one that fits. ## Why does this page not rank by success rate? Several of these endpoints have only a few measured runs, so their rates move a lot from one answer to the next. We rank by a price we can read for every provider and leave rates out. ## Do I pay while my agent is idle? No. looot bills per call, and nothing runs while your agent is idle. ## Related pages - /best/serp-api - /best/web-scraping-api - /best/news-api - /best/linkedin-api - /best/tiktok-api - /best/reddit-api - /best/b2b-data-api - /best/lead-finder - /best/email-verification-api - /best/google-maps-scraper-api - /use-cases/web-search-api - /use-cases/url-to-markdown-api - /use-cases/google-serp-api - /recipes/market-map-of-a-category ## Last reviewed 2026-10-28. The ranking by price per call is read live from the catalog on the web page. ## Agent skill for this page https://looot.ai/best/web-search-api/skill.md sets an agent up for exactly the jobs on this page, with the live price of each, and offers the page's starter task once the user has picked a job and seen the price. The line to give an agent is under "Give this to your agent" below. ## For agents ### Give this to your agent ```text set up looot for Best web search API - https://looot.ai/best/web-search-api/skill.md ```