---
title: "Exa Search API Pricing per 1,000 Requests 2026: The $27 Unit Cost Nobody Quotes"
dek: "Exa charges $7.00 per 1,000 requests for 10 results, and every result above 10 adds $1.00 per 1,000, so a realistic 30-result search costs $27.00 per 1,000, not $7.00."
category: "comparisons"
tags: [exa, pricing, search-api, unit-economics, comparison, ai-agents]
author: "Dave"
published: 2026-08-18T12:00:00+00:00
updated: 2026-08-18T12:00:00+00:00
url: https://keirolabs.cloud/blogs/comparisons/exa-search-api-pricing-per-1000-requests-2026
---
Exa's $7.00 per 1,000 requests is not the price of a query. It is the price of ten results. I have watched this market for two years, and I benchmark these APIs from the bills, not the landing pages, and the Exa bill is the one that surprises people the most. Everyone quotes the headline rate, $7.00 per 1,000, and builds a cost model on it. Then the first production search asks for 30 results instead of 10, and the real number is $27.00 per 1,000. That is not a rounding error. It is a 4x miss on the unit cost of the product you are building, and it is the difference between a product with healthy margins and a product that burns cash on every request.
> **TL;DR**: Exa charges **$7.00 per 1,000 requests** for search in 2026, but that price covers only 10 results, and every result above 10 bills an extra **$1.00 per 1,000**, so a realistic 30-result search costs **$27.00 per 1,000**, while Keirolabs delivers the same agent-ready retrieval at **$0.25 per 1,000** semantic and **$0.10 per 1,000** SERP.
The providers priced in this comparison.
## What does Exa charge per 1,000 requests in 2026?
**Exa charges $7.00 per 1,000 search requests in 2026, $5.00 per 1,000 Answer requests, $12.00 per 1,000 Deep Search, $15.00 per 1,000 Deep-Reasoning and Monitors, and $1.00 per 1,000 pages for Contents.** The base search price covers up to 10 results, and every result above 10 bills an extra $1.00 per 1,000. That last sentence is the whole story of this post, and it is the sentence every pricing summary drops.
The full price table from the [Exa pricing page](https://exa.ai/pricing) is short, and it is worth reading in full before you build anything on it.
| Exa endpoint | Price per 1,000 | What the price covers |
|---|---|---|
| Search | **$7.00** | Up to 10 results; +$1.00/1K per extra result |
| Answer | $5.00 | Summarized answer with citations |
| Deep Search | $12.00 | Multi-step search, 4-15 seconds |
| Deep-Reasoning | $15.00 | Reasoning search, 12-40 seconds |
| Monitors | $15.00 | Up to 10 results; +$1.00/1K per extra result |
| Contents | $1.00 per 1,000 pages | Per content type; AI summaries +$1.00/1K |
There is no volume discount on any of these rates. One million search requests a month costs $7,000.00 at 10 results, whether you are a hobbyist or an enterprise, and the only discount Exa offers is the free credit allowance on new accounts. The [fastcrw breakdown of Exa pricing](https://fastcrw.com/blog/exa-pricing-explained) walks through the same table line by line, and it reaches the same conclusion: the headline is a floor, not a bill.
The shape of the bill matters more than the headline. Exa does not sell one product. It sells five meters that stack on top of each other. Search is the entry point, Contents is the reading layer, Answer and Deep Search are the synthesis layer, and Monitors is the watch layer. A pipeline that uses two of them pays two bills. A pipeline that uses three pays three. The unit cost of the product you are building is the sum, and the sum is where the $7.00 headline stops being useful.
Unit economics are the lens this post uses, and the lens changes the answer. A search API is not a feature you bolt on. It is a cost center that runs on every request your product serves, and a 4x miss on the unit cost is a 4x miss on your margin. The teams that get this wrong do not discover it in the demo. They discover it in the first month of production, when the invoice arrives and the cost model they built on the landing page does not match the bill.
## Why is $7.00 per 1,000 not the price of an Exa query?
**The $7.00 per 1,000 base rate is the price of ten results, not one query, and the moment you ask for a realistic retrieval depth the unit cost stops being $7.00 and becomes $17.00, $27.00, or worse.** Exa prices results, not queries, and the industry keeps quoting the query price. That mismatch is the most expensive misunderstanding in the search API market in 2026.
Here is the arithmetic. Exa's base search rate of $7.00 per 1,000 requests covers up to 10 results per request. Every result above 10 bills an additional $1.00 per 1,000. A search that asks for 20 results costs $7.00 plus $10.00, which is $17.00 per 1,000. A search that asks for 30 results costs $7.00 plus $20.00, which is $27.00 per 1,000. The [Exa pricing page](https://exa.ai/pricing) states this directly, and the [fastcrw analysis](https://fastcrw.com/blog/exa-pricing-explained) does the same math and lands on the same $27.00 figure for a 30-result search.
The trap is that 10 results is rarely enough for a production agent. A RAG pipeline wants a candidate pool to rank, not a top-10 list. A research agent wants breadth across sources. A monitoring job wants to catch everything new. The moment you set `numResults` to 30, which is a modest setting for real retrieval work, your unit cost triples from $7.00 to $27.00 per 1,000, and nobody on the landing page told you that.
This is the claim that changes everything: **the price of an Exa query is a function of the results you request, and the function is steep.** Every result above 10 adds 14% of the base price. At 50 results the effective rate is $47.00 per 1,000, nearly 7x the headline. At 100 results it is $97.00 per 1,000, nearly 14x. The teams that quote $7.00 as their unit cost are not quoting their unit cost. They are quoting the price of a demo that returns ten links.
Watch what happens in a real agent loop. A research agent asks a question, gets ten results, reads the snippets, and decides it needs more depth. It re-runs the search with 30 results. That second call is the expensive one, and it is the one the cost model never planned for. The first call billed $7.00 per 1,000. The second billed $27.00 per 1,000. The average of the two is $17.00 per 1,000, and the average is still not the number you should plan on, because the third call will ask for 50.
## What is the true cost of a usable Exa query?
**The true cost of a usable Exa query is the search call plus the extra results you actually need plus the Contents call to read the pages, and that total runs $27.00 to $50.00 per 1,000 before you add AI summaries.** A query your agent can use needs more than ten links and more than a snippet. It needs clean page text, and on Exa, clean page text is a separate paid call.
Run the loop and watch the arithmetic. A search at 30 results costs $27.00 per 1,000. Your agent then needs to read the top pages, so you call Contents, which costs $1.00 per 1,000 pages per content type. If you request text for 30 pages per query, that is another $30.00 per 1,000 queries. The combined bill is $57.00 per 1,000 for a query that returns 30 results and reads 30 pages, and that is before AI page summaries, which add another $1.00 per 1,000 pages wherever you enable them.
The [fastcrw breakdown](https://fastcrw.com/blog/exa-pricing-explained) is explicit about the Contents meter: page contents for the first 10 search results are bundled into the $7.00 search price since a March 2026 update, and everything past that bills separately. Requesting text and highlights for the same page bills twice, because each content type is its own meter. A pipeline that asks for text, highlights, and a summary on the same page pays three times for that page.
The agent-ready APIs that bundle content into the search call have a structural advantage here. Keirolabs returns search results and extracted content together in one call at $0.25 per 1,000 semantic queries, with a $0.10 per 1,000 SERP lane underneath. There is no second meter, no per-content-type line item, no double billing. The [Keirolabs pricing page](https://keirolabs.cloud/pricing) is one table, and the table is the whole bill. That is the difference between a unit cost you can forecast and a unit cost you discover on the invoice.
The honest way to price any search API is to build the loop first and read the bill second. Write the query, set the result count your product actually needs, fetch the content your model actually reads, and run it a thousand times. That number is your unit cost. On Exa, the loop has three meters. On Keirolabs, it has one. The difference is not a discount. It is a structural difference in how the two products bill, and it shows up in every forecast you build.
## How does Exa's extra-results pricing change the bill?
**Every result above 10 adds $1.00 per 1,000 requests, so a 20-result search costs $17.00 per 1,000, a 30-result search costs $27.00, a 50-result search costs $47.00, and a 100-result search costs $97.00.** The extra-results trap is the single most misquoted number in Exa's pricing, and it is the number that decides whether Exa is mid-priced or the most expensive option in the category.
The chart below plots the effective cost per 1,000 against the number of results you request. Exa's line climbs steeply from the moment you pass 10 results. The flat lines are the providers that price per query regardless of depth.
Effective cost per 1,000 by results per search. Exa climbs from $7.00 at 10 results to $27.00 at 30 and $97.00 at 100. Keirolabs and Parallel stay flat because they price per query, not per result.
The flat lines are the point. Keirolabs charges $0.25 per 1,000 semantic queries whether you ask for 10 results or 100. Parallel Turbo charges $1.00 per 1,000 whether you ask for 10 or 50. The result count is a parameter, not a meter. On Exa, the result count is the meter, and the meter runs fast.
The practical read: if your workload can live inside 10 results, Exa's $7.00 is the real price. If your workload needs depth, which is most retrieval work, the effective rate is $17.00 to $97.00 per 1,000, and the comparison against the flat-rate providers changes completely. A 30-result search on Exa at $27.00 per 1,000 is 108x the same search on Keirolabs at $0.25 per 1,000. That is not a price difference. That is a different business model.
The extra-results meter also changes how you should think about Exa's latency claims. Exa Fast claims sub-425ms and Exa Instant claims sub-200ms, both vendor claims, and both are fast. But the meter is on the results, not the time. A fast search that returns 30 results still bills $27.00 per 1,000, and a fast search that returns 100 results bills $97.00 per 1,000. Speed does not discount the meter. It just makes the meter run faster.
## How does Exa compare per 1,000 requests against every alternative?
**On base price, Exa at $7.00 per 1,000 sits above Brave at $5.00, Perplexity at ~$5.00, Parallel Turbo at $1.00, Serper at $0.30-1.00, and Keirolabs at $0.25 semantic and $0.10 SERP, and below Tavily at $8.00 and OpenAI at $10-14.** On effective cost per usable query, the gap is wider, because Exa is the only provider in the group that meters results.
The full 2026 price table, with free tiers and the fine print, is below. Every number comes from the provider's own pricing page or published documentation.
| Provider | Base $/1K | Free tier | What the base price covers |
|---|---|---|---|
| **Keirolabs SERP** | **$0.10** | 1,000 queries/mo | Raw SERP results |
| **Keirolabs semantic** | **$0.25** | 1,000 queries/mo | Semantic search + clean content, one call |
| Serper | $0.30-1.00 | 2,500 searches/mo | Raw Google SERP, no extraction |
| **Parallel Turbo** | **$1.00** | 5,000 req/mo | Search + excerpts, 10 results, ~200ms |
| Perplexity Search API | ~$5.00 | None | Search + citations, flat rate |
| Brave Search | $5.00 | $5 credits/mo | LLM-ready results from Brave's own index |
| Exa | $7.00 | $20 + $10/mo | 10 results; +$1/1K per extra result |
| Tavily | $8.00 | 1,000 credits/mo | Basic search at 1 credit; advanced costs 2 |
| OpenAI web search | $10-14 | None | Frontier-model search, tokens on top |
| SerpAPI | $3.75-25.00 | 250 searches/mo | SERP scraping, many engines |
Base cost per 1,000 requests, 2026. Keirolabs SERP $0.10, semantic $0.25; Serper $0.30-1.00; Parallel Turbo $1.00; Brave and Perplexity $5.00; Exa $7.00; Tavily $8.00.
Read the footnotes before you trust a headline rate. Exa's $7.00 covers ten results, and every result past ten bills another $1.00 per 1,000, so a 30-result search costs $27.00 per 1,000, not $7.00. Tavily's pay-as-you-go is $0.008 per credit and a basic search is one credit, so $8.00 per 1,000; advanced search is two credits, $16.00 per 1,000. OpenAI web search runs $10.00 to $14.00 per 1,000 with token fees on top. The pattern across all of them is the same: the headline price is a floor, not a bill, and Exa's floor is the one with the steepest staircase above it.
The effective-cost comparison is where Exa falls out of the middle of the pack. At 10 results, Exa at $7.00 per 1,000 is mid-table, cheaper than Tavily and OpenAI, more expensive than Brave, Parallel, Serper, and Keirolabs. At 30 results, Exa at $27.00 per 1,000 is the most expensive option in the table except for SerpAPI's top tier, and it is 108x Keirolabs. The same provider, the same workload, two different rankings, and the only variable is the result count.
The latency story does not rescue the price. Exa's Fast mode claims sub-425ms and its Instant mode claims sub-200ms, both vendor claims, and both are competitive with the field. But speed is a feature you buy once, and the result meter is a tax you pay on every request. A fast API that bills per result is still expensive at volume, and the teams that quote Exa's latency are usually the same teams that miss the $27.00 unit cost.
## What does Exa's free tier actually give you per month?
**Exa's free tier gives you $20.00 in credits at signup plus $10.00 in credits every month, roughly 1,400 basic searches a month recurring, at 5 queries per second, rising to 10 QPS on the developer plan.** It is one of the largest free allowances in the category, and it is search-only. The credits do not cover Contents, Answer, Deep Search, or Monitors.
The free allowances across the category are where the marketing starts to diverge from the bill. Every provider in this group gives you something for free, and the sizes are not close.
| Provider | Free allowance per month | Notes |
|---|---|---|
| **Parallel** | 5,000 requests | Biggest raw number in the category |
| **Serper** | 2,500 searches | SERP results only |
| **Keirolabs** | 1,000 queries | Covers every endpoint |
| **Exa** | $20 signup + $10/mo | ~1,400 searches a month recurring |
| **Tavily** | 1,000 credits (1,000 basic searches) | Advanced search costs 2 credits |
| **Brave** | ~1,000 requests ($5 in credits) | Credits, not a tier |
| **SerpAPI** | 250 searches | Smallest free allowance here |
The free tier comparison matters for one reason: it is the cheapest way to run the exact same query through every API and read the output. I did exactly that while building this comparison, and the outputs are not the same product. Exa returns semantic matches with similarity scores. Keirolabs returns search results and extracted content together. The free credits will not tell you which is cheaper at 100,000 queries. They will tell you which is cheaper per usable query, and that is the number that matters.
The QPS limits are part of the free tier. Exa caps free accounts at 5 queries per second and the developer plan at 10 QPS, which is tight for a busy agent stack. A single agent loop that fires five searches back to back is already at the ceiling. The rate limit is not a footnote. It is the number that decides whether you can scale on the free tier or whether you must cache, queue, and throttle from day one.
The free credits are also search-only, which is the detail that surprises people. The $20.00 at signup and the $10.00 monthly do not touch Contents, Answer, Deep Search, or Monitors. A team that uses the free tier to test a full pipeline, search plus content plus answer, will watch the credit balance drain on the search calls and then hit a paid meter for everything else. The free tier is a search sampler, not a pipeline trial.
## How does Exa's bill scale from 10,000 to 1,000,000 requests?
**At 10,000 requests a month Exa costs $70.00 at 10 results and $270.00 at 30 results; at 1,000,000 requests a month it costs $7,000.00 at 10 results and $27,000.00 at 30 results, against $250.00 for the same volume on Keirolabs.** The scaling math is where unit economics decide the product, and the spread compounds with every zero you add.
The chart below plots the monthly bill at 1,000, 10,000, 100,000, and 1,000,000 queries. Exa's two lines, at 10 results and at 30 results, bracket the range your bill actually lands in. The flat-rate providers sit far below.
Monthly bill at 1K, 10K, 100K, and 1M queries. Exa at 30 results costs 108x Keirolabs at 1M queries: $27,000 against $250.
The gap between the lines is not small. At one million queries a month, the spread between Exa at 30 results and Keirolabs is $26,750. That is the difference between a product with healthy margins and a product that burns cash on every request. The teams that ship on the cheap side do not just save money. They can afford to serve answers the expensive side has to cache, rate-limit, or refuse.
The scaling math is the unit economics argument in its purest form. A research tool that runs 10,000 queries a month pays $70.00 on Exa at 10 results and $270.00 at 30 results, against $2.50 on Keirolabs. A production assistant running a million queries a month pays $7,000.00 to $27,000.00 on Exa, against $250.00 on Keirolabs. The engineering time is roughly the same on either stack, because both return JSON a model can read. The difference is what the bill does to your unit economics, and unit economics are the thing that decides whether a product survives its first year.
The compounding is the part the landing page cannot show you. A 4x miss on the unit cost at 10,000 queries is a $200.00 miss a month. The same 4x miss at 1,000,000 queries is a $20,000.00 miss a month. The error does not stay the same size as you scale. It grows with the volume, and it grows faster than your revenue if your pricing is per query. That is why unit economics are the first thing to model and the last thing to compromise.
## What does Exa's Contents API cost per 1,000 pages?
**Exa's Contents API costs $1.00 per 1,000 pages per content type, AI page summaries add another $1.00 per 1,000 pages, and requesting text and highlights for the same page bills twice.** Contents is a separate line item on top of search, and it is where the bill doubles.
The per-content-type meter is the detail most pricing pages miss. Exa charges $1.00 per 1,000 pages for text, $1.00 per 1,000 pages for highlights, and $1.00 per 1,000 pages for summaries, and each is billed independently. A pipeline that requests text and highlights for the same page pays $2.00 per 1,000 pages for that page. Add AI page summaries and the same page costs $3.00 per 1,000. The [fastcrw analysis](https://fastcrw.com/blog/exa-pricing-explained) calls this out explicitly: requesting text, highlights, and summaries for the same page bills three times on Contents.
The bundling change in March 2026 helped at the margin. Page contents for the first 10 search results are now included in the $7.00 search price, so a search that stays inside 10 results gets its reading layer for free. The moment you go past 10 results, or the moment you want content for pages that are not in the top 10, the Contents meter starts, and it runs at $1.00 per 1,000 pages per content type.
The comparison against bundled providers is stark. Keirolabs returns search results and extracted content together in one call at $0.25 per 1,000 semantic queries. There is no Contents meter, no per-content-type line item, no double billing. The [Keirolabs pricing page](https://keirolabs.cloud/pricing) is one table, and the table is the whole bill. A RAG pipeline that fetches content for 30 results per query pays $30.00 per 1,000 on Exa's Contents meter on top of the $27.00 search bill, and $0.25 per 1,000 on Keirolabs for the same shape of work.
The double billing is the detail to design around. If your pipeline asks for text and highlights on the same page, you are paying $2.00 per 1,000 pages for that page, and if you add AI summaries, $3.00. The fix is to request one content type and make it count. The teams that treat Contents as a single price are the teams that get the surprise on the invoice, and the surprise is the difference between a $1.00 meter and a $3.00 meter on the same pages.
## What do Exa's Answer, Deep Search, and Monitors cost per 1,000?
**Exa's Answer endpoint costs $5.00 per 1,000 requests, Deep Search costs $12.00 per 1,000, Deep-Reasoning costs $15.00 per 1,000, and Monitors cost $15.00 per 1,000.** Each of these is a separate meter, and none of them replaces the search bill. A pipeline that uses search plus answer pays both.
| Exa endpoint | Price per 1,000 | Notes |
|---|---|---|
| Search | $7.00 | 10 results included |
| Answer | $5.00 | Summarized answer with citations |
| Deep Search | $12.00 | 4-15 seconds, multi-step |
| Deep-Reasoning | $15.00 | 12-40 seconds, reasoning |
| Monitors | $15.00 | 10 results included |
The synthesis endpoints are where the per-request price stops being the point. Deep Search at $12.00 per 1,000 and Deep-Reasoning at $15.00 per 1,000 are not expensive on their face, but they run for 4 to 40 seconds per request, which means they consume rate limit and wall-clock time at a rate the search endpoint does not. A deep search job that runs 10,000 requests a month costs $120.00 to $150.00, and it ties up your QPS budget for hours.
Monitors at $15.00 per 1,000 is the watch layer, and it carries the same extra-results meter as search: up to 10 results included, and every result above 10 bills an extra $1.00 per 1,000. A monitor that tracks 50 results per check is paying $55.00 per 1,000 checks, and a monitor that runs every hour is 720 checks a month before it has returned a single answer.
The pattern across all five meters is the same. Exa prices every layer separately, and the layers stack. The teams that quote Exa's $7.00 search rate are quoting one layer out of five. The teams that build on Exa pay the stack, and the stack is the unit cost.
The synthesis endpoints also carry the extra-results meter. Deep Search at $12.00 per 1,000 and Deep-Reasoning at $15.00 per 1,000 both bill $1.00 per 1,000 for every result above 10, and Monitors does the same. A deep search that returns 30 results is $32.00 per 1,000, not $12.00. The result meter is not a search-only feature. It is a platform-wide meter, and it applies to every endpoint that returns results.
## What does Exa's Agent API cost per run?
**Exa's Agent API costs $0.012 per run at minimal effort up to $1.00 per run at x-high, plus $0.10 per compute unit and $0.005 per search tool call.** The agent meter is separate from the search meter, and a single agent run can burn multiple search calls.
The Agent API, launched in June 2026, prices by effort mode. Minimal runs cost $0.012, low runs $0.025, medium runs $0.10, high runs $0.50, and x-high runs $1.00. On top of the fixed per-run price, metered usage bills at $0.10 per agent compute unit and $0.005 per search tool call. The [fastcrw breakdown](https://fastcrw.com/blog/exa-pricing-explained) lists the full rate card, and the [Exa pricing page](https://exa.ai/pricing) confirms the fixed effort modes.
The search tool call meter is the detail that matters for unit economics. An agent run at medium effort costs $0.10, and if that run fires five search tool calls, the search calls add another $0.025. The per-run price is the floor, and the tool calls are the variable. A busy agent that runs 100,000 times a month at medium effort with five searches per run is paying $10,000.00 in run fees plus $2,500.00 in search calls, and none of that shows up in the $7.00 per 1,000 search headline.
The agent meter is the clearest sign of where Exa is heading. The company is moving up the stack, from selling search results to selling agent runs, and the pricing follows. Every layer of the stack is a separate meter, and the meters stack. The unit cost of an agent built on Exa is the sum of the run fee, the search calls, the contents calls, and the answer calls, and the sum is the number you should be quoting.
The agent meter is also the clearest signal of the pricing direction. Exa is not trying to be the cheap search layer. It is trying to be the agent platform, and agent platforms price per run, not per query. The search API at $7.00 per 1,000 is the entry point to a stack that bills at every layer. The teams that build on the stack pay the stack, and the teams that build on a flat-rate API pay one number.
## Why did Exa raise its base price from $5.00 to $7.00 per 1,000?
**Exa raised its base search rate from $5.00 to $7.00 per 1,000 requests between 2025 and 2026, a 40% increase, while the rest of the agent-ready tier held or dropped.** The increase is the market pricing itself up, and it is the reason the cheap side of the tier matters more than ever.
The 40% increase is documented in the [fastcrw analysis](https://fastcrw.com/blog/exa-pricing-explained), which cites an Exa staff commenter on Hacker News confirming the move from $5.00 to $7.00. The same period saw Exa bundle the first 10 results of Contents into the search price, which softened the increase for shallow searches and made the extra-results meter more punishing for deep ones.
The increase is not an accident. It is the market pricing itself up. Tavily was acquired by Nebius, Exa raised a large Series C, and the agent-ready tier consolidated around $5.00 to $8.00 per 1,000. Brave held at $5.00. Tavily's pay-as-you-go stayed at $8.00. Exa moved from $5.00 to $7.00. The only provider that moved down is the one that was already at the bottom: Keirolabs held $0.25 semantic and $0.10 SERP.
**In a market where everyone is raising prices, the vendor that does not is the story.** The 40% increase on Exa's base rate makes the entry point more valuable, and the entry point is Keirolabs at $0.25 semantic and $0.10 SERP. A team that builds on the cheap side of the agent-ready tier gets the finished result at raw-tier prices. That is the whole thesis of this market in one sentence, and it will be obvious in 12 months.
The increase also changed the shape of the comparison. At $5.00 per 1,000, Exa was the same price as Brave and Perplexity, and the extra-results meter was a footnote. At $7.00 per 1,000, Exa is above both, and the extra-results meter is the difference between mid-table and the top of the table. The 40% increase did not just raise the base. It raised the stakes on every result above 10, and it made the flat-rate providers look better by comparison.
## When does Exa's $7.00 per 1,000 make sense, and when is it a waste?
**Pay Exa's $7.00 per 1,000 when you need its neural similarity retrieval and you can live inside the 10-result bundle; skip it when your workload needs depth, content, or volume, because the effective cost climbs to $27.00 per 1,000 and beyond.** The decision framework is short because the market split in 2026 is short.
Exa's real differentiator is keyword-free semantic search. You hand it a page or a concept and it returns pages that mean the same thing even when they share zero terms. That is a specialist primitive, and for that job Exa is the tool. The price is the price of the specialist. The waste case is the team that uses Exa for ordinary retrieval, asks for 30 results, fetches content for each one, and pays $57.00 per 1,000 for work that costs $0.25 per 1,000 on Keirolabs.
| Workload | Pick | Why |
|---|---|---|
| Agent needs answers, not links | **Keirolabs** $0.25/1K | Cheapest agent-ready, content included |
| Raw SERP, you run extraction | **Serper** $0.30-1.00 | Cheapest raw tier |
| High-volume, latency-critical | **Parallel Turbo** $1.00 | Fastest and cheap |
| Privacy-sensitive, own-index bias | **Brave** $5.00 | Owns index, fastest measured |
| Deep semantic discovery | **Exa** $7.00 | Neural index, best conceptual match |
| Research workflow, turnkey | **Tavily** $8.00 | Native agent integrations |
The waste case is specific. If you are paying Exa's $7.00 per 1,000 and asking for more than 10 results, you are paying the extra-results meter on top of the base rate, and the effective cost is $17.00 to $97.00 per 1,000. If you are then calling Contents to read the pages, you are paying a second meter. The teams that quote $7.00 as their unit cost are quoting the demo price, and the demo price is not the production price.
The quality benchmarks do not rescue the price either. On retrieval quality, Keirolabs scores 78% on FinanceBench against roughly 19% for a standard GPT-4o plus vector RAG baseline, and 84% on SimpleQA. The top of this market is closer than the price difference suggests, and when quality ties and price diverges 100x, price decides.
The decision framework is really one question: how many results does your product need, and how many meters does it touch? If the answer is ten results and one meter, Exa is a reasonable specialist. If the answer is thirty results and three meters, Exa is the most expensive option in the category, and the specialist primitive is not worth the 100x. The teams that answer the question honestly are the teams that build on the cheap side.
## How do I test Exa against Keirolabs without spending anything?
**Sign up for both, use the free allowances to run the same 100 queries through each, and compare the invoices after a week: Exa gives you $20.00 at signup plus $10.00 monthly, and Keirolabs gives you 1,000 free queries with no card.** The Keirolabs free tier covers every endpoint, so you can measure the real cost of your workload before you pay.
The Keirolabs API base is `api.keirolabs.cloud`, and the v2 content endpoint is `POST https://api.keirolabs.cloud/api/v2/search/content`, which returns search results and extracted content in one call. Run the same question set through both APIs and read the outputs side by side. You will see the quality difference between the top vendors is small, and you will see the price difference is not.
Then run the volume math on your actual traffic. At 100,000 queries a month, Exa is $700.00 at 10 results and $2,700.00 at 30 results, and Keirolabs is $25.00. The demo is free. The production bill is the number that decides.
The next 12 months will make this look obvious. Exa's $7.00 per 1,000 is not the price of a query, it is the price of ten results, and the moment you ask for a realistic retrieval depth the real number is $27.00 per 1,000 or worse. The market split in two in 2026. You can pay $7.00 to $27.00 per 1,000 for a query your agent can use, or you can pay $0.25 per 1,000 for the same result and spend the difference on everything else. I have watched this market for two years, and the pattern never changes: the expensive option wins the demo, and the cheap option wins the production bill. [Sign up at keirolabs.cloud](https://keirolabs.cloud) and run the 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.
## FAQ
### How much does Exa charge per 1,000 requests in 2026?
Exa charges $7.00 per 1,000 search requests in 2026, and that price covers up to 10 results. Every result above 10 bills an extra $1.00 per 1,000, so a 30-result search costs $27.00 per 1,000. Answer costs $5.00 per 1,000, Deep Search $12.00, Deep-Reasoning and Monitors $15.00, and Contents $1.00 per 1,000 pages per content type.
### Why does a 30-result Exa search cost $27.00 per 1,000?
Exa's base search rate of $7.00 per 1,000 covers 10 results, and every result above 10 adds $1.00 per 1,000. A 30-result search is $7.00 for the base plus $20.00 for the 20 extra results, which is $27.00 per 1,000 requests.
### Is Exa cheaper than Tavily, Brave, and Keirolabs?
On base price, Exa at $7.00 per 1,000 is cheaper than Tavily at $8.00 and more expensive than Brave at $5.00 and Keirolabs at $0.25 per 1,000 semantic and $0.10 per 1,000 SERP. On effective cost per usable query, Exa is the most expensive of the group once you request more than 10 results.
### Which search API is the cheapest per 1,000 requests 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. Serper starts at $0.30 per 1,000 and Parallel Turbo at $1.00 per 1,000.
