Back to blog

    AI Development

    Best Web Search APIs for AI Agents in 2026: Pricing Compared

    By Desearch Editorial Team14 min read

    An AI agent that answers questions about the present needs a web search API: send a query over HTTP, get structured results back, and ground the model in sources it can cite. The market has split into APIs that return ranked links, APIs that return page content or excerpts sized for a context window, and APIs that return a generated answer with citations. Two long-standing options have also left the field: Microsoft retired the Bing Search APIs, and Google closed its Custom Search JSON API to new customers.

    This guide compares 11 APIs on the numbers that decide a build: the lowest self-serve price per 1,000 searches, the free tier, what a result contains, and the rate limit. Every figure comes from the vendor's own pricing page or documentation, checked on September 16, 2026, and every source is listed at the end.

    Disclosure: Desearch publishes this guide and builds one of the APIs in it. We list the same facts for every vendor and link to the official source for each so you can check them yourself.

    Quick comparison

    Prices are list prices for the lowest self-serve option that fits a standard search. They are not like-for-like: billing units, result counts, and the amount of page content differ between vendors, so read the conditions in each section before you estimate a workload.

    APILowest listed price per 1,000 searchesFree tierWhat a search returns
    Desearch$1.00 (Web Search)$5 credit when you attach a card at sign-upTitles, URLs, and snippets; AI Search adds sources across web and social platforms with an optional summary
    Brave Search API$5.00$5 in credits every month; card required for identityTitles, URLs, descriptions, optional extra snippets
    Tavily$7.50 (Project plan, basic depth)1,000 credits a month, no cardTitle, URL, content snippet, score; optional raw content and generated answer
    Exa$7.00 (up to 10 results)$20 at sign-up plus $10 a month, no payment methodTitle, URL, publish date, plus requested text, highlights, or summaries
    Parallel$1.00 (turbo or fast mode)$5 a month with a card on fileURL, title, publish date, and markdown excerpts
    Firecrawl$6.40 (Hobby, billed annually)1,000 credits a month, no cardTitles, descriptions, URLs; optional full-page markdown per result
    SerpApi$9.17 (Big Data plan)250 searches a monthParsed Google results, including organic results and SERP features
    Perplexity Search API$5.00 per 1,000 requestsNo free API allowance listedTitle, URL, snippet, date, last updated
    You.com$5.00 per 1,000 calls$100 in credits at sign-up, no cardURL, title, description, metadata; optional full page
    Linkup$5.00 (standard depth, search results)$20 at sign-up with a professional emailName, URL, content; optional sourced answer or structured output
    Google Custom Search JSON API$5.00 after 100 free queries a dayClosed to new customersTitle, link, snippet, and page metadata

    How we compared

    • Price per 1,000 searches. Where a vendor bills in credits or plans, we show the arithmetic from its published rates. We did not use promotional or enterprise pricing.
    • Free tier. The amount, whether it renews, and whether the vendor says a card is required.
    • Output. What a single search result contains, because that decides how much extraction, reranking, or summarization you still have to build.
    • Limits. Published rate limits for self-serve accounts.

    We did not run a relevance benchmark for this guide. For quality, test the same set of queries from your own application against two or three shortlisted APIs.

    1. Desearch

    Desearch offers a plain Web Search endpoint and an AI Search endpoint behind one API key, alongside dedicated X API endpoints for public posts, profiles, replies, and trends.

    • Web Search: GET /web returns ranked titles, URLs, and snippets as JSON. It is metered at $0.10 per 100 searches, which is $1.00 per 1,000.
    • AI Search: POST /desearch/ai/search answers a prompt across the sources you select (web, X, Reddit, Wikipedia, YouTube, Hacker News, and arXiv), with results you can cite and an optional final summary. It is metered at $0.40 per 1,000 items.
    • X data: X Search is metered at $0.15 per 1,000 posts.
    • Free tier: attaching a card when you sign up adds $5 in credits; usage is pay-as-you-go after that.
    • Billing visibility: billable responses report cost_usd, usage_count, and the metered service, and the same values arrive as X-Desearch-* headers.

    Best for: agents that need web results and public X data from one integration, and teams that meter usage per request.

    2. Brave Search API

    Brave serves results from its own independent index, which it describes as over 30 billion pages.

    • Price: $5 per 1,000 requests on the Search plan. The Answers plan is $4 per 1,000 requests plus $5 per million input and output tokens.
    • Free tier: $5 in free credits every month. Brave asks for a card to confirm identity and states it will not charge it on the free plan.
    • Output: GET /res/v1/web/search returns titles, URLs, and descriptions, with up to five extra snippets per result on request. Controls include country, language, freshness, safesearch, count (up to 20), and offset.
    • Limits: the Search plan lists 50 queries per second.

    Best for: results that do not depend on Google or Bing, with locale and freshness controls on every query. See Desearch vs Brave Search API.

    3. Tavily

    Tavily builds search for LLM workflows, with optional raw page content and a generated answer.

    • Price: a basic search costs 1 credit and an advanced search costs 2. Pay-as-you-go is $0.008 per credit ($8.00 per 1,000 basic searches). The Project plan is $30 for 4,000 credits, or $0.0075 per credit ($7.50 per 1,000 basic searches).
    • Free tier: 1,000 API credits a month with no credit card; credits reset on the first day of each month.
    • Output: POST /search returns a title, URL, content snippet, and relevance score for each result, plus cleaned page content when include_raw_content is true and an optional LLM-generated answer.
    • Limits: 100 requests per minute on development keys and 1,000 on production keys.

    Best for: RAG pipelines that want snippets, raw content, and an optional answer from one call. See Desearch vs Tavily.

    4. Exa

    Exa offers several search types, from standard search to multi-step deep search.

    • Price: /search is $7 per 1,000 requests for up to 10 results. Each result beyond 10 is $1 per 1,000, and AI page summaries are $1 per 1,000 pages. Deep search types cost $12 to $15 per 1,000.
    • Free tier: $20 in credits at sign-up and $10 in credits every month, with no payment method required.
    • Output: each result carries a title, URL, and publication date, plus the contents you request, such as text, highlights, or a summary.
    • Limits: 10 queries per second for /search, and 5 for deep search types.

    Best for: retrieval where you want to choose how much page content comes back. See Desearch vs Exa.

    5. Parallel

    Parallel's Search API takes a natural-language objective and returns excerpts formatted for LLMs.

    • Price: turbo and fast modes are $1 per 1,000 requests with 10 results by default; basic and advanced modes are $5 per 1,000. Parallel's docs note that requests default to advanced when no mode is set. Extra results are $1 per 1,000.
    • Free tier: $5 in free credits each month for organizations with a credit card on file.
    • Output: POST /v1/search returns a URL, title, publish date, and markdown excerpts for each result. Full page content comes from a separate Extract API.
    • Limits: 600 search requests per minute.

    Best for: agents that describe what they need in plain language and want compact excerpts. Set the mode explicitly, because the default changes the price fivefold.

    6. Firecrawl

    Firecrawl combines search with its scraping stack, so a search can return full-page content for each result.

    • Price: a search costs 2 credits per 10 results, and scraping a result adds 1 credit per page. The Hobby plan is $19 a month billed monthly or $16 billed annually for 5,000 credits, which works out to $7.60 or $6.40 per 1,000 searches of up to 10 results. Firecrawl marks these rates as effective September 4, 2026.
    • Free tier: 1,000 credits a month with no credit card, which Firecrawl equates to 500 searches.
    • Output: POST /v2/search returns titles, descriptions, URLs, and query-relevant highlights; add scrapeOptions for markdown, HTML, links, or screenshots of each result.
    • Limits: 10 search requests per minute on the Free plan and 100 on Hobby.

    Best for: workflows that search and then need the full content of each page. See Desearch vs Firecrawl.

    7. SerpApi

    SerpApi parses results from Google and other search engines, including SERP features.

    • Price: monthly plans. Starter is $25 for 1,000 searches ($25.00 per 1,000); Developer is $75 for 5,000 ($15.00); Production is $150 for 15,000 ($10.00); Big Data is $275 for 30,000 ($9.17). Only successful searches count toward the allowance.
    • Free tier: 250 searches a month.
    • Output: organic results with position, title, link, displayed link, and snippet, plus other SERP elements, with a Markdown output option for LLMs.
    • Limits: hourly throughput by plan, from 50 searches an hour on Free to 6,000 on Big Data.

    Best for: applications that need Google's result layout, rankings, or SERP features rather than just sources. See Desearch vs SerpApi.

    8. Perplexity Search API

    Perplexity's Search API returns raw results without a generated answer; answers come from its separate Agent API.

    • Price: $5 per 1,000 requests. A request can carry up to five queries and still counts as one billing unit, with no token charges. Invalid and rate-limited requests are not billed.
    • Free tier: the API docs describe prepaid credits and do not list a free API allowance.
    • Output: POST /search returns a title, URL, snippet, date, and last-updated date for each result, with 1 to 20 results per query.
    • Limits: 50 query units per second.

    Best for: agents that fan one task out into several related queries. See Desearch vs Perplexity API.

    9. You.com

    You.com's Web Search API returns web and news results in one request.

    • Price: $5 per 1,000 calls, with 1 to 100 results per call. Live full-page extraction adds $1 per 1,000 pages; pages served from cache are free.
    • Free tier: $100 in complimentary credits at sign-up with no credit card, plus a keyless option that allows 100 free queries a day.
    • Output: each result has a URL, title, description, and metadata such as publication date; set extraction_mode to full_page for markdown or HTML of the whole page.
    • Limits: 10 requests per second on self-serve accounts.

    Best for: combined web and news results, with full pages available when you need them.

    10. Linkup

    Linkup lets you choose between ranked results, a sourced answer, and structured JSON from the same endpoint.

    • Price: standard-depth search results are $0.005 per call ($5.00 per 1,000); sourced answers and structured output are $0.006 ($6.00 per 1,000). Deep search is $0.05 per call for results ($50.00 per 1,000).
    • Free tier: $20 in credits when you sign up with a professional email address, topped back up to $20 each month for eligible accounts.
    • Output: POST /v1/search returns a name, URL, and content for each result, or an answer with inline citations, or JSON matching a schema you supply.
    • Limits: 10 queries per second per organization.

    Best for: agents that want structured output or a cited answer without a separate generation step.

    11. Google Custom Search JSON API

    Google's documentation now states that the Custom Search JSON API is closed to new customers and that existing customers have until January 1, 2027 to move to an alternative.

    • Price for existing customers: 100 free queries a day, then $5 per 1,000 queries, up to 10,000 queries a day.
    • Output: each item includes a title, link, snippet, and page metadata, with up to 10 results per request and 100 results in total.
    • Scope: new Programmable Search Engines must use a list of sites to search, and Google suggests Vertex AI Search, which it is renaming Agent Search, for searching up to 50 domains.

    If you run Custom Search today, see the Google Custom Search API alternative migration map.

    What happened to the Bing Search API?

    Microsoft retired the Bing Search APIs on August 11, 2025, including Bing Custom Search, Spell Check, Autosuggest, and Visual Search. Microsoft pointed customers to Grounding with Bing Search in Azure AI Agents, which supplies web data to an agent rather than returning results to your own code. The Bing Search API alternative page maps each Bing call to a replacement.

    How to choose

    • You need web results and public X data together: Desearch covers both from one key.
    • You want an index that is independent of Google and Bing: Brave Search API.
    • You need full page content with each result: Firecrawl, You.com with full-page extraction, or Tavily with raw content.
    • You need Google's result layout and SERP features: SerpApi.
    • You want a cited answer or structured output from the search call: Linkup, Tavily, or Desearch AI Search.
    • The lowest list price per 1,000 plain searches: Desearch Web Search and Parallel in turbo or fast mode are both listed at $1.00.
    • You are replacing Bing or Google Custom Search: start with a migration map, then run your existing queries against two candidates.

    Whatever you shortlist, measure a complete task rather than a single call: the searches, any extracted pages, and any generated answer your application needs.

    Frequently asked questions

    Is there a free web search API?

    Several APIs include free usage. Tavily and Firecrawl each give 1,000 credits a month without a card, Exa gives $20 at sign-up plus $10 a month without a payment method, SerpApi includes 250 searches a month, and You.com starts accounts with $100 in credits. Brave and Parallel give $5 a month but ask for a card, and Desearch adds $5 in credits when you attach a card at sign-up.

    What is the cheapest web search API?

    At list price for plain searches, Desearch Web Search ($0.10 per 100 searches) and Parallel in turbo or fast mode ($1 per 1,000 requests) are the lowest in this comparison at $1.00 per 1,000. Brave, Perplexity, You.com, and Linkup list $5.00 per 1,000 for their standard search. Result counts and content differ, so compare the cost of a full task.

    Is the Google Custom Search API free?

    It includes 100 free queries a day for existing customers and charges $5 per 1,000 queries after that, up to 10,000 a day. Google has closed the API to new customers, and existing customers have until January 1, 2027 to transition.

    What replaced the Bing Search API?

    Microsoft retired the Bing Search APIs on August 11, 2025 and pointed customers to Grounding with Bing Search in Azure AI Agents. Applications that called the REST API and parsed results themselves typically move to another web search API instead.

    Sources

    All pages checked on September 16, 2026.

    Put your next idea to work.

    Try a query in the playground, then use the docs to bring Desearch into your application.

    Get API keyRead docs
    Browse more articles