WebSearch
APIs
Submit an API

The guide · updated Oct 2026

How to choose a search API.

Ignore the leaderboards. Five questions decide it, and you probably know three of the answers already.

Which one should you buy?

Three questions. Your shortlist updates as you answer.

1Who reads the results?
2What do you need back?
3What matters most?

Your shortlist

  1. Answer the three questions and your top five shows up here.

Scored from the directory's data. Treat it as a starting point, then run your own bake-off.

The five questions

01

Who reads the results?

If a person reads them, you want fast, well-ranked links and snippets, so any independent index or SERP API will do. If a model reads them, the snippet is rarely enough. You'll end up fetching every page, so pick an API that does the fetching and cleaning for you, or plan to build that yourself.

02

What do you need back?

The four shapes are links, page text, a written answer and Google's exact results page. They're different products, so don't compare their prices as if they were the same thing. A $1 SERP call plus your own scraper can easily cost more than a $8 call that returns clean text.

03

Whose index is it?

This is the risk question. Microsoft retired the Bing Search APIs in August 2025. Google's Custom Search stopped taking new customers, and its whole-web search ends on January 1, 2027. Google's suit against SerpApi was dismissed in July 2026, but without prejudice, and Reddit's separate suit is still pending. Scraped-Google APIs work today, and they also carry risk that isn't on the invoice.

04

What does it cost at your volume?

Work it out per useful result, not per call. Count retries, the extra fetches, and the tokens the results add to every later turn of an agent loop. Built-in model tools look cheap until those tokens show up.

05

What can you do with the results?

Read the terms before you cache anything. Brave, for example, only allows storing results for training or tuning on plans that explicitly grant storage rights. Built-in search from model vendors ties you to their models. Yandex has different terms for non-residents of Russia and Kazakhstan.

Prices per 1,000 searches

APISearch, per 1,000Details
Keirolabs ↗$0.03–5.33search/lite · 0.0375 credits $0.10 · Google SERP · 2 credits $5.33 · search/fast · 0.5 credits $1.33 · search/content (page text) · 1.5 credits $4. Paid-plan credits per call: lite 0.0375, fast 0.5, content 1.5, answer 1, Google SERP 2, deep 2. Credits cost $2.67/1k on Essential ($29.99 for 11,250) and $0.89/1k on Startup ($99.99 for 112,500).
Serper ↗$0.30–1Google SERP $1. Prepaid credits: $1.00/1k on the $50 pack down to $0.30/1k on the $3,750 pack. Credits expire after 6 months.
Parallel ↗$1–5turbo $1 · fast $1 · basic $5 · advanced $5. Search includes 10 results; extra results $1/1k. Same price on every volume.
Brave ↗$5web search $5. $5 of free credits every month. Prepaid, no subscription.
Linkup ↗$5–50flash / fast / standard $5 · deep $50. Prepaid balance, topped back up to $20 free each month. Deep search is 10× standard.
Perplexity ↗$1–5fast $1 · standard $5. No token charges on the Search API.
You.com ↗$5web search $5. Page contents cost $1/1k pages extra.
Exa ↗$4–12lite $4 · auto $7 · deep $12. 10 results included; extra results $1/1k. Page contents $1/1k pages extra.
Tavily ↗$5–16ultra-fast / fast / basic $8 · advanced $16. $0.008/credit pay-as-you-go, $0.005 on the Growth plan. Basic, fast and ultra-fast are 1 credit; advanced is 2.
SerpApi ↗$1.96–25Google SERP $25. Monthly plans: $25 for 1,000 searches, falling to about $2/1k on the largest Cloud plans.
OpenAI Web ↗$10–25 + tokensweb search tool $10 · web search preview (non-reasoning) $25. Plus search content tokens at model rates. The preview tool on non-reasoning models is $25/1k with free content tokens.
Claude Web ↗$10 + tokensweb search tool $10. Plus token costs. Results count as input tokens on every later turn.

Official list prices, checked on each vendor's pricing page on November 2, 2026. Click a name for its source. Work out your monthly bill →

Then run your own bake-off.

Every comparison you'll find online was written by someone selling a search API. Some of them are honest. None of them used your queries. Spend an afternoon on this.

  1. Take 50–100 real queries from your logs or users, not made-up ones.
  2. Run them through 2–3 shortlisted APIs on the same day, from the same region.
  3. Grade answerability: is the answer anywhere in the top 5 results?
  4. Measure p50 and p95 latency while running calls in parallel, not one at a time.
  5. Count downstream tokens, the cost that hides in your model bill.
  6. Compute cost per useful result. That's your number.

Sources