Expert Guide

Best Search APIs for AI Agents 2026: Cost per 1k Queries

Nine web search APIs for agents and RAG compared on price per 1,000 queries, free monthly allowance and what each call returns, with a costed month of 50,000 searches.

Product links may be affiliate links. How we rate 9 tools compared
TL;DR

Short answer: Brave Search API is the best default for most agents in 2026, at $5 per 1,000 requests from its own independent index, with a free credit every month. Exa suits semantic retrieval for RAG at $4 to $7 per 1,000 searches, Tavily is the quickest to wire into an agent framework, and Serper and Parallel are the cheapest at $1 per 1,000. Prices checked on vendor pages on 21 October 2026.

At 50,000 searches a month, the bill runs from $50 on Serper, Parallel Fast or Perplexity Fast Search to $725 on SerpApi. Most of that spread is about what comes back: a Google results page, ranked excerpts, or page text a model can read without a second call.

Key facts9 tools compared: Brave Search API, Exa, Tavily, Parallel Search API, Perplexity Search API…
  • Updated: October 24, 2026
  • Top pick: Brave Search API (best for: Agents that need fresh web results from an independent index at one flat per-request price)
  • Top pick price as of October 24, 2026: Brave Search API: Search $5 per 1,000 requests plus a monthly free credit; Answers $4 per 1,000 plus token fees
  • 9 tools compared: Brave Search API, Exa, Tavily, Parallel Search API, Perplexity Search API, Linkup, Firecrawl, Serper, SerpApi
  • Exa (best for: RAG pipelines and research agents that want semantic search and page text from one vendor): Lite $4, Auto $7, Deep $12 per 1,000 searches (up to 10 results); $10 free credit every month
  • Tavily (best for: Agent builders who want search, extract and crawl behind one credit balance with a free monthly tier): 1,000 free credits a month; Project $30/mo for 4,000 credits; pay-as-you-go $0.008 per credit
  • Parallel Search API (best for: High-volume agent loops that need sub-second results at the lowest per-request price here): Turbo and Fast $1, Basic and Advanced $5 per 1,000 requests (10 results); 5,000 free requests a month

A search API for an agent is judged on three things the headline price hides: whose index answers the query, how much page text comes back, and what happens when the free allowance runs out. A cheap results page that forces a second scrape call on every hit is not cheap. This page prices each API per 1,000 searches and for one month of agent traffic, using figures read on each vendor's own pricing page or docs on 21 October 2026.

Toolradar data: Toolradar, the software directory we run, evaluated 486 products for its October 2026 API tools ranking, and 131 of the 200 it ranks list a free plan. Web search APIs built for agents are a small corner of that category, so the order below rests on per-request prices and output format, not on the directory score.

Methodology: nine APIs that return live web results to code, ordered by fit for agent and RAG work. We weigh index ownership, output a model can use without a second call, price per 1,000 searches and the size of the free monthly allowance. There is no hands-on benchmark here; every figure comes from the vendor. Pricing moves in this category are the kind of change the Devshot daily brief flags.

Top Picks

Based on features, real-world fit, and value for money.

#1 Top pick

Brave Search API

Every result comes from Brave's own index with no fallback to another engine, at one flat rate and 50 queries per second.

Pricing

Search $5 per 1,000 requests plus a monthly free credit; Answers $4 per 1,000 plus token fees

Best for

Agents that need fresh web results from an independent index at one flat per-request price

#2

Exa

An AI-native search engine with web, people, company and scholarly indexes, plus page contents billed per 1,000 pages.

Pricing

Lite $4, Auto $7, Deep $12 per 1,000 searches (up to 10 results); $10 free credit every month

Best for

RAG pipelines and research agents that want semantic search and page text from one vendor

#3

Tavily

A credit-based search API for agents, where a basic search costs 1 credit and an advanced search costs 2.

Pricing

1,000 free credits a month; Project $30/mo for 4,000 credits; pay-as-you-go $0.008 per credit

Best for

Agent builders who want search, extract and crawl behind one credit balance with a free monthly tier

Best Search APIs for AI Agents 2026: Cost per 1k Queries: 9 tools compared, updated Oct 2026
ToolPricingBest for
Brave Search APISearch $5 per 1,000 requests plus a monthly free credit; Answers $4 per 1,000 plus token feesAgents that need fresh web results from an independent index at one flat per-request price
ExaLite $4, Auto $7, Deep $12 per 1,000 searches (up to 10 results); $10 free credit every monthRAG pipelines and research agents that want semantic search and page text from one vendor
Tavily1,000 free credits a month; Project $30/mo for 4,000 credits; pay-as-you-go $0.008 per creditAgent builders who want search, extract and crawl behind one credit balance with a free monthly tier
Parallel Search APITurbo and Fast $1, Basic and Advanced $5 per 1,000 requests (10 results); 5,000 free requests a monthHigh-volume agent loops that need sub-second results at the lowest per-request price here
Perplexity Search APISearch API $5 per 1,000 requests, or $1 per 1,000 with Fast Search; no token costsTeams that want raw web results with filtering, or a search tool inside Perplexity's Agent API
LinkupStandard search $0.005 per call, deep search ten times that; $20 monthly top-up for eligible accountsAgents that want a sourced answer or structured output from one search call, paid from a prepaid balance
FirecrawlSearch 2 credits per 10 results; 1,000 free credits a month; Hobby $16/mo billed annually for 5,000 creditsAgents that search and then scrape the result pages, paid from one shared credit balance
SerperPrepaid packs from 50,000 Google searches at $1.00 per 1,000, down to $0.30 per 1,000; 2,500 free queriesCost-first agents that only need Google results pages and fetch page text elsewhere
SerpApiFree plan 250 searches a month; Starter $25/mo for 1,000; Production $150/mo for 15,000 searchesAgents that need structured results from Google plus Bing, Amazon and other engines through one API

Pricing read from each vendor's own pricing page, checked Oct 2026. Every vendor here publishes a price.

Lowest published monthly price by toolBar chart of the lowest monthly price each vendor publishes; the same figures are in the comparison table above.Lowest published monthly priceTavily$30Firecrawl$16SerpApi$25
Lowest monthly figure each vendor publishes, checked Oct 2026. A tilde marks a figure the vendor states approximately. Per-seat and usage charges can sit on top of it. 6 of 9 do not publish a comparable monthly price and are left out rather than estimated.
1 Brave Search API logo
Top Pick

Brave Search API

Every result comes from Brave's own index with no fallback to another engine, at one flat rate and 50 queries per second.

Pricing
Search $5 per 1,000 requests plus a monthly free credit; Answers $4 per 1,000 plus token fees

Best for: Agents that need fresh web results from an independent index at one flat per-request price

Pros

  • Search returns full results plus extra context formatted for LLMs, at up to 50 queries per second
  • The monthly free credit covers 1,000 Search requests, enough for a prototype agent
  • Goggles let you rerank or filter results with your own rules, such as dropping low-quality domains

Cons

  • A credit card is required at sign-up, even to use the free monthly credit
  • The Answers plan is capped at 2 queries per second, and its per-token fee comes on top of the request price
2 Exa logo

Exa

An AI-native search engine with web, people, company and scholarly indexes, plus page contents billed per 1,000 pages.

Pricing
Lite $4, Auto $7, Deep $12 per 1,000 searches (up to 10 results); $10 free credit every month

Best for: RAG pipelines and research agents that want semantic search and page text from one vendor

Pros

  • Three search modes trade cost for depth, from Lite for tight agent loops to Deep for research
  • The Contents endpoint returns page text for $1 per 1,000 pages, so a separate scraper is often unnecessary
  • The monthly free credit buys 2,500 Lite searches, and an MCP server is included

Cons

  • Each result above 10 adds to the per-1,000 price, and AI page summaries are billed on top
  • The free tier is capped at 10 search queries per second; pay-as-you-go raises it to 25
3 Tavily logo

Tavily

A credit-based search API for agents, where a basic search costs 1 credit and an advanced search costs 2.

Pricing
1,000 free credits a month; Project $30/mo for 4,000 credits; pay-as-you-go $0.008 per credit

Best for: Agent builders who want search, extract and crawl behind one credit balance with a free monthly tier

Pros

  • The free Researcher plan renews 1,000 credits on the first of each month, with no card required
  • Extract bills 1 credit per 5 successful URLs, and a failed extraction costs nothing
  • Monthly plans fall from $0.0075 to $0.005 per credit as volume grows

Cons

  • Basic search costs $7.50 per 1,000 on Project and $8 on pay-as-you-go, above most rivals here
  • Advanced search doubles the credit cost, and free-plan requests stop once the monthly credits run out
4 Parallel Search API logo

Parallel Search API

Four search processors from about 200ms to about 3 seconds, returning ranked URLs with compressed excerpts.

Pricing
Turbo and Fast $1, Basic and Advanced $5 per 1,000 requests (10 results); 5,000 free requests a month

Best for: High-volume agent loops that need sub-second results at the lowest per-request price here

Pros

  • Turbo returns in about 200ms and Fast in about 700ms, both at the lowest per-request rate on this page
  • Its free monthly allowance is the largest on this list, enough for steady development traffic
  • Excerpts come back compressed and ranked, sized for a model's context window

Cons

  • Results beyond the first 10 are billed per 1,000 extra results, on every processor
  • Full page text is a separate Extract API, billed per 1,000 results
5 Perplexity Search API logo

Perplexity Search API

Raw ranked web results billed per successful request, with a Fast Search mode at a fifth of the standard rate.

Pricing
Search API $5 per 1,000 requests, or $1 per 1,000 with Fast Search; no token costs

Best for: Teams that want raw web results with filtering, or a search tool inside Perplexity's Agent API

Pros

  • A successful request that carries several queries is billed once, with no token charge
  • Fast Search cuts the rate to a fifth for latency-sensitive calls
  • Inside the Agent API, the web_search tool costs $2.50 per 1,000 calls

Cons

  • Agent API runs add model tokens to every tool call, so the cost depends on the model you pick
  • Reading a page is a separate fetch_url tool call at $0.0005 each
6 Linkup logo

Linkup

Prepaid per-call search with an answer mode priced close to raw results and a deep mode at ten times the standard rate.

Pricing
Standard search $0.005 per call, deep search ten times that; $20 monthly top-up for eligible accounts

Best for: Agents that want a sourced answer or structured output from one search call, paid from a prepaid balance

Pros

  • A sourced answer or structured output costs $6 per 1,000, one dollar more than raw results
  • Accounts that sign up with a professional email are topped back up to $20 each month
  • You can pay per request in USDC through the x402 protocol, with no account, at a $0.01 minimum per request

Cons

  • Deep search costs ten times the standard rate, so one setting can move the bill a lot
  • When the prepaid balance runs out, the API returns a 429 error rather than billing overage
7 Firecrawl logo

Firecrawl

A scraping platform whose search endpoint draws on the same credits as its scrape and crawl calls.

Pricing
Search 2 credits per 10 results; 1,000 free credits a month; Hobby $16/mo billed annually for 5,000 credits

Best for: Agents that search and then scrape the result pages, paid from one shared credit balance

Pros

  • Search and scrape share one balance: 2 credits per 10 search results, 1 credit per scraped page
  • Standard is $83/mo billed annually for 100,000 credits, enough for 50,000 ten-result searches
  • A scrape that returns no result is not charged

Cons

  • Credits do not roll over on Hobby, Standard or Growth; only Scale carries them for one month
  • Hobby is $19/mo on monthly billing, and a page that answers 403 or 404 still costs 1 credit
8 Serper logo

Serper

Prepaid Google search results with no subscription, and the lowest per-1,000 price on this list at high volume.

Pricing
Prepaid packs from 50,000 Google searches at $1.00 per 1,000, down to $0.30 per 1,000; 2,500 free queries

Best for: Cost-first agents that only need Google results pages and fetch page text elsewhere

Pros

  • The $375 pack buys 500,000 searches at $0.75 per 1,000, and the largest pack reaches $0.30
  • Credits are prepaid with no subscription, and a 7-day refund applies if under 20% is used
  • Results usually return in 1 to 2 seconds, with country and language set per query

Cons

  • It returns Google results pages, so text for a prompt needs a separate scrape call
  • Credits expire after 6 months, so a large pack can lapse before a slow project uses it
9 SerpApi logo

SerpApi

The broadest engine coverage on this page, sold as monthly plans that make it the most expensive per search.

Pricing
Free plan 250 searches a month; Starter $25/mo for 1,000; Production $150/mo for 15,000 searches

Best for: Agents that need structured results from Google plus Bing, Amazon and other engines through one API

Pros

  • Covers Google, Bing, Amazon and more engines, each through its own endpoint
  • Only successful searches count; cached, errored and failed searches are free
  • The number of results in a response does not change the cost of the search

Cons

  • Starter works out to $25 per 1,000 searches, the highest entry rate on this page
  • Hourly throughput caps apply, 200 an hour on Starter and 1,000 on Developer

What it is

A search API for AI agents is a web search endpoint that returns ranked results, snippets or page text as JSON, so a model can ground its answer in current pages instead of its training data.

Why it matters

The per-request rate is only half the bill. At 50,000 searches a month, Brave, Linkup and Perplexity's standard Search API tie in the middle of the table, Exa costs $200 on Lite and $350 on Auto, and Tavily's basic search costs $316. That Tavily figure is our arithmetic on its printed prices: the Startup plan's 38,000 credits for $220, then 12,000 pay-as-you-go credits at $0.008 each.

The free tiers differ by a factor of twenty once you convert them into searches. Parallel prints 5,000 free requests a month, and Linkup's monthly top-up for professional-email accounts covers about 4,000 standard searches. Exa's monthly credit buys 2,500 Lite searches, Brave's credit and Tavily's 1,000 credits each cover 1,000, Firecrawl's 1,000 credits cover 500 ten-result searches, and SerpApi's free plan gives 250. A prototype under 1,000 searches a month runs free on most of this list. For the model bill that sits beside the search bill, see best LLM APIs and free LLM APIs.

Key features to look for

Index ownershipEssential
Brave answers every query from its own crawl, with no fallback to another engine. Serper returns Google results pages, and SerpApi returns pages from Google, Bing and other engines. An independent index matters when you need predictable results and no third party in the data path.
Output a model can readEssential
A results page with snippets needs a scrape step before the model sees the text. Exa, Parallel, Linkup and Firecrawl can return page text or excerpts in the same product, which removes a call per result.
Price per extra result
Exa and Parallel price the first 10 results per request, then charge per 1,000 extra results. An agent that asks for 25 results per query pays more than the headline rate.
Latency modes
Parallel lists Turbo at about 200ms and Advanced at about 3 seconds. Exa lists 450ms to 1 second. Fast modes cost less on Parallel and Perplexity, so an interactive agent can be the cheaper workload.
Credit expiry and rollover
Serper credits last 6 months. Tavily credits reset on the first of the month. Firecrawl rolls credits over only on Scale. Match the expiry rule to how steady your traffic is.

Pricing

Methodology: on 21 October 2026 we read each vendor's own pricing page or billing docs and kept the figures they print. The pricing table below costs one workload, 50,000 searches in a month with up to 10 results each, on the cheapest published path for each API. Usage-based prices are monthly by nature; Firecrawl's plan figure is billed annually, because that is the price it prints for Standard, and monthly billing costs more. Free allowances are left out of the totals. How we weigh a page is on how we rate.

Brave's API page prints the Search and Answers plans, each with a monthly free credit, and says no query falls back to another engine. Exa's pricing lists Lite, Auto and Deep search per 1,000 requests and the Contents endpoint per 1,000 pages. The Tavily credits page lists Project, Bootstrap at $100 for 15,000 credits, Startup and Growth at $500; the Tavily pricing page shows the same free tier. Toolradar keeps a profile of Tavily and of Parallel Web Systems.

Parallel's pricing splits Search into four processors at two rates. Perplexity's API pricing gives the Search API and Fast Search rates and the Agent API tool fees. Linkup's billing docs list each depth and output type, and the Linkup pricing page states the free tier as a query count.

Firecrawl's pricing sets search at 2 credits per 10 results; our Firecrawl review covers the scraping side, and best web scraping tools compares it with dedicated scrapers. Serper's site prints four prepaid packs from Starter to Ultimate. SerpApi's pricing runs from Starter through Big Data at $275 for 30,000 searches to Searcher at $725 for 100,000. Both have Toolradar profiles: Serper and SerpApi, and Firecrawl has one too.

PlanPriceBest for
Brave Search API (Search plan)$250 a monthFull results and LLM context from Brave's own index; the free credit covers the first 1,000
Exa Lite search$200 a monthLowest-latency Exa mode with up to 10 results per request
Exa Auto search$350 a monthExa's default mode; page text through Contents is billed separately
Tavily basic search (Startup plan)$316 a month38,000 plan credits for $220, then 12,000 pay-as-you-go credits
Tavily advanced search (Growth plan)$500 a monthAdvanced search uses 2 credits per call, so this volume needs the Growth plan's full allowance
Parallel Search (Fast or Turbo)$50 a monthRanked URLs with compressed excerpts in under a second
Parallel Search (Basic or Advanced)$250 a monthLonger excerpts per source and multi-hop research
Perplexity Search API with Fast Search$50 a monthLower-latency raw web results, no token costs
Perplexity Search API (standard)$250 a monthRaw web results with advanced filtering
Linkup standard search (raw results)$250 a monthPrepaid balance; eligible accounts get a monthly top-up
Linkup standard search (sourced answer)$300 a monthAn LLM composes a cited answer from the results
Firecrawl Standard$83/mo billed annually100,000 credits, all spent on searches at this volume, none left for scraping
Serper Starter pack$50 one-time50,000 Google search credits valid for 6 months, no subscription
SerpApi Searcher$725 a monthSmallest plan that covers 50,000 searches; it includes 100,000
Mistakes to avoid
×Comparing a results-page API with a page-text API on price alone. Serper returns Google results, so each hit your agent reads still needs a scrape call, which a search-plus-contents API covers in one product.
×Asking for more than 10 results by default. Exa and Parallel bill every extra result per 1,000, so a 25-result query costs more than the headline rate.
×Leaving Tavily on advanced search. It costs 2 credits instead of 1, which doubles the bill for the same number of searches.
×Buying a large Serper pack for a slow project. The credits expire after 6 months, so unused searches are lost.
×Budgeting Firecrawl search without the scraping that follows. Both draw on one balance, so a plan sized for searches alone runs dry once the agent reads the pages.
Expert tips
→Count searches per task before you pick a vendor. An agent that runs five searches per user question turns 10,000 questions into 50,000 searches, the volume priced in the table.
→Use the cheap fast modes for the inner loop and the deep modes only for research steps. Parallel and Perplexity both price their fast search at a fifth of the standard rate.
→Stack free tiers while you prototype. Parallel's free requests, Linkup's monthly top-up and Exa's monthly credit together cover most development traffic, so you can compare result quality on real queries before a card is charged.
→Expose the search API to your agent through MCP when the vendor ships a server. See <a href="/reviews/best-mcp-servers-for-developers">best MCP servers for developers</a> for the wider list.

The bottom line

Start with Brave Search API when you want an independent index at one flat rate, and add Exa when retrieval quality for RAG matters more than cost per call. Pick Tavily if your agent framework already ships its tool and you stay near the free tier, and move to Parallel Fast or Serper once volume makes the lowest per-1,000 rate the number that matters. Agents that write code against these APIs are covered in best AI coding agents; vector storage for the retrieved text is in best free vector databases; and model routing through one gateway is in OpenRouter alternatives.

Check the free allowance and the expiry rule before you commit, because they change the real bill more than the headline rate. For proxy-based scraping at scale, see our Bright Data review. Cite this: Devshot, "Best Search APIs for AI Agents 2026: Cost per 1k Queries", October 2026.

Frequently asked questions

What is the best web search API for AI agents in 2026?

Brave Search API is the best default for most agents: one flat per-request rate from its own independent index, with a free credit every month, per Brave's API page on 21 October 2026. Exa is the stronger pick for semantic retrieval in RAG, while Serper and Parallel Fast win when the price per search decides.

What is the cheapest search API for AI agents?

Serper, Parallel's Fast and Turbo processors and Perplexity's Fast Search tie for the lowest entry rate, $1 per 1,000 searches. Serper drops to $0.30 per 1,000 on its largest prepaid pack, the lowest rate on this page, but it returns Google results pages rather than page text.

Is there a free web search API for agents?

Several APIs renew a free allowance every month. Parallel gives 5,000 free requests, Linkup tops eligible accounts up to $20 (about 4,000 standard searches), Exa gives $10 of credit, Tavily gives 1,000 credits and SerpApi gives 250 searches. A prototype can run on these allowances before any paid plan starts.

How much does Tavily cost per search?

A basic Tavily search costs 1 credit and an advanced search costs 2. Credits cost $0.008 each on pay-as-you-go and $0.0075 to $0.005 on monthly plans, from Project at $30 for 4,000 credits up to Growth, per Tavily's credits page in October 2026. The free plan includes 1,000 credits a month.

Exa vs Tavily: which is cheaper?

Exa is cheaper per search at most volumes. Exa Lite costs $4 per 1,000 searches and Auto $7, while Tavily's basic search costs $7.50 per 1,000 on its $30 Project plan and $8 on pay-as-you-go. At 50,000 searches a month, Exa Lite is $200 and Tavily basic is $316 on the Startup plan plus overage.

Do I need a scraper as well as a search API?

Only if your search API returns results pages rather than page text. Serper and SerpApi return search results, so reading a page needs a scrape call. Exa Contents, Parallel Extract, Linkup Fetch and Firecrawl scrape cover that step, each billed separately per page or per call.
Related guides

Some offers on this page may be paid placements or contain affiliate links.

Get the Devshot brief

Free daily newsletter, read in 5 minutes.

Subscribe free