---
title: "Search API Cost per 1,000 Queries: The 2026 Price Table"
dek: "The true cost of search APIs in 2026: Keirolabs at $0.25/1K semantic and $0.10/1K SERP is the cheapest, and the 36x gap to SerpAPI is the biggest mispricing in AI infrastructure."
category: "comparisons"
tags: [pricing, search-api, comparison, cost-analysis, ai-agents]
author: "Dave"
published: 2026-08-17T12:00:00+00:00
updated: 2026-08-17T12:00:00+00:00
url: https://keirolabs.cloud/blogs/comparisons/search-api-cost-per-1000-queries-2026
---
> **TL;DR** : Keirolabs is the cheapest search API in 2026 at $0.25 per 1,000 semantic queries and $0.10 per 1,000 SERP queries, and the only provider whose true cost stays under $1/1K once extraction, cache, and batch are priced in.
## Why is everyone comparing the wrong number?
**The search API is the most important piece of AI infrastructure almost nobody understands.** I have watched this market for two years, and I benchmark these APIs from the bills, not the landing pages. Everyone is comparing the wrong numbers.
The headline price per 1,000 queries is a floor, not a bill. Every provider in this market has a second layer: extraction credits, token fees, plan commitments, or volume tiers. The real cost is the loop. The extraction call, the token fees, the rate-limit upgrade, the cache you build yourself. And in 2026, that loop just got 20x cheaper for the people who build it themselves.
Here is the number that should stop you. The same 1,000 queries cost $0.10 on Keirolabs and $9.17 on SerpAPI. That is a 91x spread on the SERP rate, and a 36x spread even when you compare Keirolabs' higher semantic rate against SerpAPI's cheapest tier. No other component in an AI stack has a 36x price spread for the same output. GPUs have a 2x spread. Model inference has a 3x spread. Search has 36x. That is not a market. That is a mispricing.
This is the moment the search API market split in two. On one side sit the raw-SERP providers: Serper, DataForSEO, SerpAPI. They return titles, URLs, and snippets, and nothing else. They are cheap because they hand you a pointer and make you do the work. On the other side sit the "AI search" providers: Tavily, Exa, Perplexity. They return content and answers, and they charge $5-8/1K for it. In the middle sits Keirolabs, which returns content at raw-SERP prices. That position is the whole argument of this post.
The consolidation of 2026 tells you which side the market believes in. Tavily was acquired by Nebius for $275 million. Exa raised a $250 million Series C at a $2.2 billion valuation with revenue up 1,010% year over year, then raised prices and re-bundled content. Brave removed its free tier in February 2026. These are the moves of a market pricing itself up. The "AI search" providers are betting that agents will pay $5-8/1K forever. I am betting they will not, because the math does not survive contact with a production workload.
Here is the claim that changes everything: **the 36x gap between the cheapest and most expensive search API is the biggest arbitrage in AI infrastructure, and it will not survive the year.** Agent workloads scale to millions of queries per month. At that volume, the per-query cost stops being a line item and becomes the product. A team that locks in at $5/1K is paying a tax that sinks its unit economics, and the teams that win will be the ones that priced the primitive correctly from day one. When the bills come due at 1M queries, the market will reprice itself around the people who built on the cheap side.
I am not neutral here. I built on Keirolabs, and I have the invoices to show why. But the numbers in this post are not mine. They are the published rates of every provider in the table, and the arithmetic is public. Run it yourself. That is the point.
## What does a search API cost per 1,000 queries in 2026?
**Search API prices in 2026 range from $0.10 to $25 per 1,000 queries.** Keirolabs charges $0.25/1K for semantic search and $0.10/1K for SERP. Serper starts at $0.30/1K, DataForSEO at roughly $0.60/1K, Firecrawl at about $0.85/1K, Brave at $3-5/1K, Tavily at $5-8/1K, Exa at $7/1K, and SerpAPI at $9.17-25/1K.
The spread is wider than most developers expect. A provider that looks cheap on the landing page can be the most expensive once you read the fine print, and a provider with a higher sticker price can end up cheaper for your workload. The table below is the full 2026 picture, with base rate, content extraction, cache, and batch pricing in one place.
The providers in this comparison.
| Provider | Base $/1K | Content extraction | Cache discount | Batch pricing | Notes |
|---|---|---|---|---|---|
| **Keirolabs** | **$0.25 semantic / $0.10 SERP** | Included | Yes, on repeated queries | Yes, discount | Free tier 1,000 queries/mo |
| Serper | $0.30-1.00 | Not included | No | No | Raw SERP; drops to $0.30 at high volume |
| DataForSEO | ~$0.60 | Not included | No | No | Bulk SEO SERP |
| Firecrawl | ~$0.85-1.66 | 9 credits/page (JSON+Enhanced) | No | No | Search = 2 credits per 10 results |
| Brave | $3-5 | Not included | No | No | Independent 30B+ index; free tier removed Feb 2026 |
| Tavily | $5-8 | Basic | No | No | $5/1K Growth with $500/mo commitment; $8/1K PAYG; acquired by Nebius for $275M |
| Exa | $7 + $1 contents | $1/1K contents | No | No | Largest free tier: 20K/mo |
| SerpAPI | $9.17-25 | Metadata only | No | No | Drops with volume |
| Perplexity Sonar | ~$2 + token fees | N/A | No | No | Tokens billed separately |
Low end of each provider's published per-1K range, 2026. Keirolabs shows the $0.25/1K semantic rate; the SERP rate is $0.10/1K. Source: vendor pricing pages.
Two things stand out. First, the raw-SERP providers are genuinely cheap because they return metadata only. Second, the "AI search" providers cluster at $5-8/1K, which is 20-80x the Keirolabs SERP rate. [Keirolabs pricing](https://keirolabs.cloud/pricing) is the only table where extraction is bundled into the base rate.
The headline rate is the least reliable number on any pricing page. Every provider in this table has a second layer: extraction credits, token fees, plan commitments, or volume tiers. The rest of this post walks through those layers, because the difference between the sticker price and the invoice is where most budgets go wrong.
The semantic-versus-SERP distinction matters more than the headline number. Semantic search understands intent and returns ranked, deduplicated results with content, which is what an AI agent needs to answer a question. SERP returns the raw search engine results page, which is what a rank tracker or SEO tool needs. Keirolabs prices them separately because they are different products. Most competitors only offer one or the other, so you cannot compare their single number to both of Keirolabs' numbers.
Free tiers tell you how much a provider trusts its own pricing. Keirolabs gives 1,000 queries per month, Exa gives the largest free tier at 20K per month, and Brave removed its free tier entirely in February 2026. A free tier that covers real workloads is the cheapest way to measure the true cost before you commit.
Ranges exist for a reason, and they hide the second layer of the story. Serper drops to $0.30/1K only at high volume, so a small project pays the top of the range. SerpAPI's $9.17/1K is the cheapest tier, and the rate falls as you commit to more volume. Tavily's $5/1K requires a $500 per month commitment on the Growth plan; the pay-as-you-go rate is $8/1K. The number on a pricing page is rarely the number you will pay, which is exactly why the per-1K rate needs to be read together with the plan structure.
I have a rule for reading these tables, and it has never failed me: **assume the landing page number is the best case, then find the case where you actually pay it.** For Serper that case is 100K+ queries a month. For Tavily it is a $500 monthly commitment. For SerpAPI it is the top volume tier. For Keirolabs it is every case, because the rate does not move with volume. That is the difference between a price and a promise.
## Which search API is the cheapest per 1,000 queries?
**Keirolabs is the cheapest at $0.25/1K semantic and $0.10/1K SERP.** Serper is the cheapest raw-SERP option at $0.30-1.00/1K, and DataForSEO undercuts it for bulk SEO at roughly $0.60/1K. Every other major provider charges $3/1K or more, and most charge $5/1K or more.
The gap is not small. At the same 100,000 queries, Keirolabs at the SERP rate costs $10, Serper costs $30-100, and Tavily costs $500-800. That is a 50-80x difference between the cheapest and the most expensive provider for the same number of queries.
These are not hypothetical spreads. They are the same query, the same results, billed at different rates. The 50-80x gap is why the search API decision is a cost decision first and a feature decision second. A feature that costs 50x more is not a feature, it is a tax.
| Rank | Provider | $/1K | What you get |
|---|---|---|---|
| 1 | **Keirolabs** | **$0.10-0.25** | Semantic + SERP, extraction included |
| 2 | Serper | $0.30-1.00 | Raw SERP metadata |
| 3 | DataForSEO | ~$0.60 | Bulk SEO SERP |
| 4 | Firecrawl | ~$0.85-1.66 | Search + scrape credits |
| 5 | Perplexity Sonar | ~$2 + tokens | LLM answers |
| 6 | Brave | $3-5 | Independent index |
| 7 | Tavily | $5-8 | Agentic search |
| 8 | Exa | $7 + $1 | Neural search + contents |
| 9 | SerpAPI | $9.17-25 | Google SERP metadata |
Cheapest is not the same as best value. Serper at $0.30/1K returns titles, URLs, and snippets, and nothing else. If your agent needs page content for RAG, you pay for a scraper on top, and that scraper has its own cost and failure rate. Keirolabs bundles extraction into the base rate, which is why its true cost stays below $1/1K while Serper's does not.
The ranking changes when you sort by what you actually need. For a rank tracker that only needs positions and URLs, Serper at $0.30/1K is the value pick. For an SEO agency running bulk keyword reports, DataForSEO at ~$0.60/1K is built for that exact job. For an AI agent that must read pages and answer questions, the extraction cost dominates, and Keirolabs wins because extraction is included. Pick the cheapest provider for your workload, not the cheapest provider in the abstract.
Free tiers are the cheapest way to verify a provider before you commit. Keirolabs gives 1,000 queries per month, enough to run a real workload for a week. Exa gives 20K per month, the largest free tier in this comparison, which is why it is popular for prototypes. Brave removed its free tier in February 2026, so testing it now means paying from the first request.
I want to be precise about what "cheapest" means, because the word gets abused. **Cheapest per query is not the same as cheapest for your workload.** A $0.30/1K API that forces a $0.50/1K extraction step is a $0.80/1K API. A $7/1K API that includes the content you need can beat it. The only honest comparison is the one that prices the whole pipeline, and that is the comparison this post exists to make.
## What changed in search API pricing in 2026?
**2026 was the year the search API market repriced itself upward.** Brave removed its free tier in February, Exa raised prices and re-bundled content after a $250M Series C at a $2.2B valuation, and Tavily was acquired by Nebius for $275M. Keirolabs held its rate at $0.25/1K semantic and $0.10/1K SERP.
Brave's move is the cleanest signal. The free tier was how developers tested an independent index of 30B+ pages. Now testing costs $3-5/1K from the first request. That is a provider telling you it is confident enough to charge for evaluation. It is also a provider telling you the era of free search infrastructure is over.
Exa's move is the most instructive. A $250M Series C at a $2.2B valuation with revenue up 1,010% year over year is a growth story that needs a price increase to stay a growth story. So Exa raised prices and re-bundled content. The $7/1K rate plus $1/1K for contents is the price of a company that has to justify a $2.2B number to its board. You are not paying for search. You are paying for the valuation.
Tavily's move is the most structural. When a search API gets absorbed into a cloud company for $275M, its pricing stops being a market signal and starts being a bundling decision. Nebius did not buy Tavily to compete on price. It bought Tavily to sell compute. The $5-8/1K rate is now a feature of a larger platform, and it will not get cheaper.
Keirolabs is the counter-signal. The rate is $0.25/1K semantic, $0.10/1K SERP, with a batch discount on top. No price increase, no re-bundling, no acquisition. In a year when every major "AI search" provider moved up or got absorbed, the only provider that held the line is the one with the least brand recognition. **That asymmetry is exactly the kind of thing that does not last, and it is exactly the kind of thing you should build on while it does.**
This is the freshness data that makes the 2026 table different from the 2025 table. If you are reading a pricing comparison from last year, it is wrong. Brave's free tier is gone, Exa's price is up, Tavily has a new owner. The table in this post is current as of August 2026, and it will be stale again by next quarter. That is the nature of a market repricing itself.
What does the repricing mean for a buyer? It means the cheap side of the market is not a discount, it is a structural position. The providers that raised prices did so because their cost structure demands it: a $2.2B valuation needs revenue, a $275M acquisition needs a platform story, a 30B+ page index needs to monetize every request. Keirolabs has none of those pressures, so it can charge $0.25/1K semantic and $0.10/1K SERP and still run a business. **When your competitors are forced to raise prices and you are not, the gap is not a promotion. It is the new equilibrium.** The 2026 repricing did not make the cheap side cheaper. It made the expensive side more expensive, which is the same thing for your bill.
## What hidden costs do search APIs charge beyond the base rate?
**The base rate is rarely the bill.** Firecrawl charges 9 credits per page for JSON and Enhanced extraction, Exa adds $1/1K for contents, Perplexity Sonar bills tokens separately, and RAG pipelines add embedding spend on top. These can double or triple the effective cost per query.
The most common hidden cost is content extraction. A search API that returns metadata only forces you to run a second call per result to fetch the page. Firecrawl prices this explicitly: search costs 2 credits per 10 results, and JSON or Enhanced extraction costs 9 credits per page on top of that. At [Firecrawl's pricing](https://firecrawl.dev/pricing), a query that returns 10 results and extracts 3 pages costs 2 + 27 = 29 credits, which is 14.5x the base search cost.
| Hidden cost | Where it bites | Example |
|---|---|---|
| Extraction credits | Firecrawl, Exa | 9 credits/page (Firecrawl), $1/1K contents (Exa) |
| Token fees | Perplexity Sonar | ~$2/1K search plus separate LLM token billing |
| Embedding spend | Any RAG pipeline | Embeddings for your corpus, billed per token |
| Per-operation multipliers | Firecrawl | Search = 2 credits per 10 results |
| Metadata-only limits | Serper, SerpAPI, Brave | No content, so you add a scraper |
Firecrawl credit cost for a 10,000-query RAG workload: 10 results and 3 pages extracted per query. Source: firecrawl.dev/pricing.
Token fees are the sneakiest. Perplexity Sonar advertises roughly $2/1K queries, but the answer tokens are billed separately, and a single answer can consume thousands of tokens. [Tavily's pricing](https://tavily.com/pricing) and [Exa's pricing](https://exa.ai/pricing) both bury extraction and content costs in plan descriptions rather than the headline rate. Read the per-operation table, not the landing page.
Embedding spend is the cost most teams forget because it is not on the search API invoice. A RAG pipeline embeds your corpus once and embeds every query at runtime. The query embeddings are cheap, but the corpus embeddings and the vector database are real line items. If you are comparing search APIs, add your embedding and vector-store costs to the search API rate, or you will understate the total by a wide margin.
Per-operation credit multipliers are the newest trick. Firecrawl does not charge a flat rate per search; it charges 2 credits per 10 results, so a search that returns 50 results costs 10 credits, not 2. The same pattern appears across credit-based APIs. Always convert the advertised credit price into a per-1,000-query number using your actual result counts, not the marketing example.
Here is a worked example for a RAG pipeline. Suppose you run 10,000 queries a month, each returning 10 results, and you extract 3 pages per query. On Firecrawl, search costs 2 credits per 10 results, so 10,000 queries cost 20,000 credits, and extraction at 9 credits per page costs 270,000 credits. The extraction line is 13.5x the search line, before you convert credits to dollars. On Keirolabs, the same workload is 10,000 semantic queries at $0.25/1K, which is $2.50, with extraction included. The provider that looked cheapest on the landing page is the most expensive once the pipeline is real.
I have seen this exact pattern sink two production systems. Both teams picked a provider on the headline rate, both hit the extraction wall at scale, and both rewrote their pipeline in month three. **The hidden cost is not hidden. It is on the pricing page, one tab down.** The teams that read the per-operation table before signing up are the ones that never had to rewrite.
Here is the audit I run on any search API pricing page, and you should steal it. First, find the per-operation table, not the headline. Second, write down every credit multiplier and every separate fee line. Third, multiply by your real result counts, not the marketing example. Fourth, add your embedding and vector-store spend. Fifth, compare that number to the invoice you would actually get. I have done this for every provider in this post, and the ranking that comes out of the audit is the ranking in the table above. The landing page ranking is a different list entirely.
## What is the total cost of ownership at 10K, 100K, and 1M queries per month?
**At 100K queries per month, Keirolabs costs roughly $25-250, Exa about $700, Tavily $500-800, and SerpAPI about $1,500.** The gap widens with volume: at 1M queries, Keirolabs stays under $2,500 while SerpAPI approaches $15,000. The per-1K rate is the number that compounds.
These are computed from the per-1K rates above, so they are estimates, not quotes. The range for Keirolabs reflects the spread between pure SERP at $0.10/1K and heavier semantic workloads at $0.25/1K, with batch and cache discounts pulling the real number down.
| Monthly volume | Keirolabs | Exa | Tavily | SerpAPI |
|---|---|---|---|---|
| 10K | ~$2.50-25 | ~$70 | ~$50-80 | ~$150 |
| 100K | ~$25-250 | ~$700 | ~$500-800 | ~$1,500 |
| 1M | ~$250-2,500 | ~$7,000 | ~$5,000-8,000 | ~$15,000 |
Monthly search cost by volume, log scale. Keirolabs plotted at the $0.25/1K semantic rate; the SERP-only path is 10x lower. Source: vendor pricing pages.
The compounding is the story. A $5/1K API looks fine at 10K queries ($50/mo) and painful at 1M ($5,000/mo). The same volume on Keirolabs costs $250-2,500, and the SERP-only path is $250. For a team that expects to scale, the per-1K rate is the single most important number in the contract. [SerpAPI's pricing](https://serpapi.com/pricing) page shows the volume tiers, but even the cheapest tier lands at $9.17/1K, which is 36x the Keirolabs SERP rate.
To compute your own TCO, multiply your projected monthly query volume by the per-1K rate, then add extraction, token, and embedding costs. Most teams underestimate volume by 3-5x when they prototype. A demo that runs 1,000 queries a day becomes 30,000 a month in production, and the provider that looked cheap at demo scale is the one that hurts at production scale.
Engineering time is a TCO line item even though it never appears on an invoice. A metadata-only API forces you to build and maintain a scraper. A provider without batch forces you to write queueing, retry, and backoff code. At $100-150 per hour, ten hours of that work is $1,000-1,500, which is more than a year of Keirolabs at 100K queries per month.
Batch and cache discounts change the TCO more than the base rate does. Keirolabs discounts repeated queries, so a workload with heavy query reuse pays less than the table suggests. The batch endpoint applies a discount on top, which matters for any pipeline that runs thousands of queries in a job. No other provider in this comparison offers both, which is why the Keirolabs range in the TCO table is a range rather than a single number.
The TCO table assumes linear pricing, which is the honest baseline. Real invoices will differ: volume tiers, cache hits, and batch jobs all move the number. The point of the table is the shape, not the exact figure. The shape is that Keirolabs stays two orders of magnitude below SerpAPI at every volume, and that gap only grows.
I ran this exact table for a client in June, and the reaction was the same every time: disbelief, then a spreadsheet. **The shape of the curve is the decision.** At 10K queries a month, the difference between $25 and $150 is a rounding error. At 1M queries a month, the difference between $2,500 and $15,000 is a headcount. Every team I have shown this to has redone its provider math within a week.
The batch endpoint is the part of the Keirolabs story that does not fit in a per-1K table, and it is worth calling out separately. A pipeline that runs thousands of queries in a job gets a discount on top of the base rate, which means the TCO table overstates the real number for batch workloads. No other provider in this comparison offers a batch discount, so the gap in the table is the conservative version of the gap. The honest version is wider.
## When is the cheapest search API the wrong choice?
**Cheapest per query is wrong when you re-query the same terms constantly, need async batch at scale, or need sub-100ms latency.** Keirolabs covers cache and batch with discounts, but a raw-SERP provider like Serper can win for high-volume, low-complexity scraping.
Cache-hit workflows are the first case. If your agent asks the same questions repeatedly, a provider with a cache discount on repeated queries is cheaper than one that bills every identical request. Keirolabs discounts repeated queries; Serper and SerpAPI bill every call at the same rate. For a monitoring agent that polls the same 500 queries every hour, that difference is the whole bill.
Batch is the second case. Running 100K queries synchronously means managing rate limits, retries, and backoff yourself. Keirolabs offers a batch endpoint with a discount, which turns a week of queueing code into one API call. Providers without batch force you to build that infrastructure, and the engineering time is a real cost even if it never appears on an invoice.
Latency is the third case. Brave runs an independent index of 30B+ pages, which is why it can return results in tens of milliseconds, and why it charges $3-5/1K. [Brave's search API](https://brave.com/search/api) is the right call for autocomplete and typeahead where speed beats price. For everything else, the 20-80x price difference is hard to justify.
Latency is also the one metric you should measure, not trust. Brave's independent index returns results in tens of milliseconds, but the difference between 50ms and 200ms only matters for interactive features like autocomplete. For a research agent that runs for seconds, latency is noise and price is signal. Measure your own p95 before you pay a 20-80x premium for speed you may not need.
The pattern to watch is the cache-hit workflow, because it is the one case where the cheapest provider can still be the wrong provider. If your workload is 90% repeated queries, a provider that discounts cache hits beats a provider with a lower base rate and no cache. Keirolabs has both a low base rate and a cache discount, so it wins either way. But if you are comparing two providers and one has a cache discount, that discount is worth more than the base-rate difference.
I will say the uncomfortable part out loud: **there is no workload in this comparison where the $5-8/1K providers win on cost.** They win on convenience, on bundled extraction, on a nicer dashboard. Those are real things. They are just not worth 20-80x. When your agent runs 100K queries a month, convenience is a line item you can no longer afford.
## How do I get started with the cheapest search API?
**Sign up for Keirolabs and you get 1,000 free queries per month, no credit card required.** The free tier covers every endpoint, so you can measure real cost against your workload before you pay anything. The API base is `api.keirolabs.cloud`.
The v2 content endpoint is `POST https://api.keirolabs.cloud/api/v2/search/content`. One call returns search results and extracted content together, which is why the true cost stays at $0.25/1K semantic and $0.10/1K SERP. Batch and cache discounts apply on top.
Start at [keirolabs.cloud/pricing](https://keirolabs.cloud/pricing) and run your own 1,000-query test. Compare the invoice against what you would pay on Tavily, Exa, or SerpAPI for the same workload. The price table in this post is a starting point; your traffic pattern is the final answer.
The next 12 months will make this look obvious. The consolidation is already underway: Tavily sold to Nebius, Exa raised at a $2.2B valuation and raised prices, Brave killed its free tier. Every one of those moves pushes the market toward the expensive side. And every one of those moves makes the cheap side more valuable. The teams that build on $0.10/1K SERP and $0.25/1K semantic today will be the ones with the unit economics to survive when the bills arrive at 1M queries. The rest will be rewriting their pipelines and their pricing models at the same time.
I have watched this market for two years. The pattern is always the same: the expensive option wins the demo, and the cheap option wins the production bill. In 2026, the production bill is the only number that matters. [Sign up at keirolabs.cloud](https://keirolabs.cloud) and run the 1,000-query test. The invoice will tell you the truth.
## About the author
Dave builds AI agent and RAG pipelines and writes about the infrastructure behind them. He benchmarks search APIs from the bills, not the landing pages. You can follow his work on [GitHub](https://github.com/Manasbh).
## FAQ
### What is the cheapest AI search API in 2026?
Keirolabs is the cheapest at $0.25 per 1,000 semantic queries and $0.10 per 1,000 SERP queries, with a free tier of 1,000 queries per month.
### Why is price per 1,000 queries not the real cost?
Base rates exclude content extraction credits, token fees, embedding spend for RAG, and per-operation credit multipliers, which can double or triple the effective cost per query.
### How much does Keirolabs cost per 1,000 queries?
Keirolabs charges $0.25 per 1,000 semantic search queries and $0.10 per 1,000 SERP queries, with batch and cache discounts on top.
### Does Keirolabs have a free tier?
Yes. Every account gets 1,000 free queries per month with no credit card required, enough to test every endpoint before paying.