---
title: "Exa Search API Pricing 2026: The $7/1K Headline Is the Most Misleading Number on the Page"
dek: "Exa Search API pricing in 2026 is a stack of per-endpoint rates, per-content-type charges, and an extra-results surcharge that can quadruple the $7/1K headline to $27/1K."
category: "comparisons"
tags: [exa, pricing, search-api, comparison, cost-analysis, 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-2026
---
Exa's $7.00 per 1,000 requests is the cheapest-looking number on its pricing page, and it is the most misleading number in the search API market. I have watched this market for two years, and I benchmark these APIs from the bills, not the landing pages. The real Exa bill is built from per-endpoint rates, per-content-type charges, and an extra-results surcharge that can quadruple the headline price. Everyone is comparing the wrong number, and it is costing them real money.
This is the moment the search API market split in two. On one side you have the flat-rate vendors who sell a finished, agent-ready result for a number you can budget. On the other side you have Exa, which sells a primitive and then charges you for every way you try to use it. The pricing page shows $7.00 and the invoice shows a stack. The headline rate is the bait. The per-endpoint rates, the per-content-type charges, and the extra-results surcharge are the bill. In 12 months this will be obvious, and the teams that read the invoice now will be the ones still standing.
The extra-results surcharge changes everything about how you should read an Exa invoice, because it means the headline rate is not a rate at all. It is the price of a 10-result search, and every result past the tenth is a separate line item. A 30-result search costs $27.00 per 1,000, not $7.00. That is not a rounding error. It is a 3.9x multiple on the number everyone quotes, and it is sitting in plain sight on the pricing page.
I am not here to tell you Exa is bad. Exa built the best keyword-free semantic search engine in the world, and the find-similar primitive is genuinely unmatched. I am here to tell you that the industry is comparing the wrong numbers, and the mistake is costing teams real money. The teams that quote Exa's $7.00 as their unit cost are not quoting their unit cost. They are quoting the price of one step out of four, and the other three steps arrive on the invoice with different names.
> **TL;DR**: Exa Search API pricing in 2026 is a stack: $7.00/1K search covers only 10 results, every result above 10 bills +$1.00/1K (a 30-result search is $27/1K), Contents costs $1.00/1K per content type, and the Agent API runs $0.012 to $1.00 per run, so the real bill is 2x to 4x the headline.
The providers priced in this comparison.
## What does the Exa Search API cost per 1,000 requests in 2026?
**Exa Search API charges $7.00 per 1,000 search requests, $5.00 per 1,000 answer requests, $12.00 per 1,000 deep search requests, $15.00 per 1,000 deep-reasoning and monitor requests, and $1.00 per 1,000 pages on Contents, with every endpoint billed separately.** That is the full price table on the [Exa pricing page](https://exa.ai/pricing), and it is the first thing most people miss: there is no single Exa price. There are six prices, and the one you pay depends on which endpoint your agent calls and how you configure it.
| Exa endpoint | Price | What it does |
|---|---|---|
| Search | **$7.00 / 1,000 requests** | Semantic web search, up to 10 results |
| Answer | **$5.00 / 1,000 requests** | LLM-generated answer grounded on search |
| Deep Search | **$12.00 / 1,000 requests** | Multi-step research over many queries |
| Deep-Reasoning Search | **$15.00 / 1,000 requests** | Deep research with a reasoning loop |
| Monitors | **$15.00 / 1,000 requests** | Scheduled re-runs of a query |
| Contents | **$1.00 / 1,000 pages** | Page content, billed per content type |
The endpoint split is the whole game. A naive agent that calls Search and then calls Contents to read the pages is paying two vendors' worth of line items on one invoice. A research agent that calls Deep Search is paying $12.00 per 1,000, nearly double the headline. A monitoring setup that re-runs a query on a schedule is paying $15.00 per 1,000 on top of the search that feeds it. The headline $7.00 is the cheapest endpoint on the page, and it is the one that appears in every blog post and every comparison table, including the ones that should know better.
The plans layer on top of the endpoints. Exa's free tier runs at 5 queries per second, the developer plan at 10 QPS, and the enterprise tier goes higher with custom terms. The QPS number matters more than most teams think, because a 5 QPS ceiling turns a real-time agent into a queue. A product serving a few hundred concurrent users needs more than 5 queries a second before it has shipped, and the jump from free to developer is where the capacity conversation starts. The [Exa pricing page](https://exa.ai/pricing) is the source for all of it, and the [FastCrw breakdown](https://fastcrw.com/blog/exa-pricing-explained) walks through the same table line by line.
Exa endpoint prices, 2026. The $7.00 search headline is the second-cheapest endpoint on the page, and every other endpoint is billed on top of it.
## Does the Exa Search API have a free tier in 2026?
**Exa gives you $20 in credits at signup plus $10 in credits every month, which works out to roughly 1,400 basic searches a month recurring, at 5 QPS on the free tier and 10 QPS on the developer plan.** It is a real free allowance, not a trial, and it is one of the more generous ones in the category. The catch is what the credits buy. A basic search at $7.00 per 1,000 is $0.007 per query, so $10 in monthly credits covers about 1,400 searches before the meter starts.
| 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, no card |
| **Exa** | $20 signup + $10/mo | ~1,400 basic searches recurring |
| **Brave** | ~1,000 requests ($5 in credits) | Credits, not a tier |
| **Tavily** | 1,000 credits (1,000 basic searches) | Advanced search costs 2 credits |
| **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. The outputs are not the same product, and the free allowances 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 credits also expire on a rolling basis, which is the detail most pricing pages miss. The $10 that lands every month does not accumulate forever. It is a monthly allowance, not a balance, and the moment you stop building, the allowance stops mattering. For a prototype it is plenty. For a production agent it is a rounding error, because a production agent does not run 1,400 searches a month. It runs 1,400 searches a day, and the free tier covers the first 45 minutes.
The free tier also caps you at 5 QPS, which is the second constraint hiding under the credits. Even with unlimited credits, 5 queries a second is a single-user prototype, not a product. The developer plan doubles that to 10 QPS, and the enterprise tier goes higher with custom terms. The QPS ladder is part of the pricing, and the teams that ignore it discover the ceiling the same way they discover the surcharge: on the day a traffic spike hits.
## How does Exa bill per content type on the Contents endpoint?
**Exa's Contents endpoint charges $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.** This is the line item that quietly doubles the bill. Search returns links. Contents returns the page. If you want the text and the highlights, you pay for both, because each content type is a separate charge.
| Contents request | Price per 1,000 pages |
|---|---|
| Text | **$1.00** |
| Highlights | **$1.00** |
| AI page summary | **+$1.00** (on top of text) |
| Text + highlights, same page | **$2.00** (billed twice) |
The math is brutal for RAG pipelines. A search that returns 10 results, followed by a Contents call for text on all 10 pages, is $0.007 for the search plus $0.01 for the contents, and that is before you ask for anything else. Add AI summaries at +$1.00 per 1,000 pages and the contents line doubles again. A pipeline that wants text, highlights, and summaries for the same 10 pages is paying $0.03 in contents charges for a search that cost $0.007. The [FastCrw breakdown](https://fastcrw.com/blog/exa-pricing-explained) calls this out explicitly: the per-content-type model is where Exa's real margin lives.
The contrast with the flat-rate vendors is stark. Keirolabs returns search results and extracted content together in a single call at $0.25 per 1,000 semantic queries, with a $0.10 per 1,000 SERP lane underneath. There is no per-content-type meter, because the content is part of the search. Exa sells the search and the content as separate products, and the invoice reflects it. The teams that build on the flat-rate side do not just save money. They save the engineering time of stitching two endpoints together, and they save the surprise when the contents line arrives on the bill.
## How much does Exa's extra-results surcharge add to a search?
**Exa's $7.00 per 1,000 search requests covers up to 10 results, and every result above 10 bills +$1.00 per 1,000, so a 30-result search costs $27.00 per 1,000, not $7.00.** This is the surcharge that can quadruple the headline price, and it is the most under-reported number on the page. The $7.00 is not the price of a search. It is the price of a 10-result search. Every result past the tenth is a separate line item.
| Results per search | Cost per 1,000 searches |
|---|---|
| 10 | **$7.00** |
| 20 | **$17.00** |
| 30 | **$27.00** |
| 50 | **$47.00** |
The arithmetic is simple and the consequence is not. A 30-result search is 3x the results and 3.9x the price. A 50-result search is 5x the results and 6.7x the price. The surcharge scales faster than the value, because the first 10 results are the ones that matter for most queries and the next 40 are the ones that get filtered out by your reranker anyway. Teams that set a high result count to improve recall are paying for results they discard.
Effective cost per 1,000 Exa searches. A 30-result search is 3x the results and 3.9x the price, because every result above 10 bills +$1.00 per 1,000.
The surcharge also changes the comparison math against every other vendor. Parallel Turbo returns 10 results for $1.00 per 1,000, flat. Brave returns up to five snippets for $5.00 per 1,000, flat. Keirolabs returns search results and content for $0.25 per 1,000, flat. None of them meter the result count. Exa is the only vendor in the category that charges you more for asking for more results, and it is the only one that needs a footnote on its headline rate.
## What does the Exa Agent API cost per run?
**Exa's Agent API charges $0.012 per run at minimal effort up to $1.00 per run at x-high, plus $0.10 per compute unit, $0.005 per search tool call, $0.02 per email enrichment, and $0.07 per phone enrichment, with the default effort:auto floating anywhere inside that range.** This is the cost lottery. The headline range is an 83x spread, and the default setting does not tell you where in the range you land until the run finishes.
| Agent API component | Price |
|---|---|
| Run, minimal effort | **$0.012** |
| Run, x-high effort | **$1.00** |
| Compute unit | **$0.10** |
| Search tool call | **$0.005** |
| Email enrichment | **$0.02** |
| Phone enrichment | **$0.07** |
The effort parameter is the lottery ticket. Set effort to minimal and a run costs $0.012. Set it to x-high and the same run costs $1.00, an 83x difference on the same task. The default, effort:auto, lets the model decide, which means the price of a run is a function of the query, the model's mood, and the number of tool calls it decides to make. A run that makes 20 search tool calls adds $0.10 in search charges on top of the run price, and a run that enriches a phone number adds $0.07 for a single field. The [Exa pricing page](https://exa.ai/pricing) lists the components, and the [FastCrw breakdown](https://fastcrw.com/blog/exa-pricing-explained) does the math on a realistic run.
The unpredictability is the product. A flat-rate search API gives you a number you can put in a spreadsheet. The Agent API gives you a range, and the range is wide enough to hide a 10x cost overrun. A team that budgets $0.02 per run and ships a workload that lands at $0.20 per run has a 10x surprise on its hands, and the surprise arrives on the invoice, not at the design review. The teams that build on flat-rate search do not have this problem, because the price is the price.
The compute unit is the second dial. Every run burns compute units at $0.10 each, and the number of units a run burns is not published in advance. Add the search tool calls at $0.005 each and the enrichments at $0.02 to $0.07 per field, and a single agent run can touch five line items before it returns a result. The components are individually small. The stack is not. This is the same shape as the search endpoint, just with a wider range and a model deciding the price.
## Why did Exa raise its base search price from $5 to $7?
**Exa raised its base search rate from $5.00 to $7.00 per 1,000 between 2025 and 2026, a 40% increase on the headline rate, while the rest of the stack got cheaper.** The price history matters because it tells you which direction the vendor is heading. Exa's base search was $5.00 per 1,000 in 2025. It is $7.00 per 1,000 in 2026. That is a 40% increase in a single year, on the cheapest endpoint, while the vendors below it held their lines or cut prices.
The increase is not an accident. Exa runs a 144-H200-GPU cluster it calls the Exacluster, and it sells semantic retrieval from a custom vector database. That infrastructure is expensive, and the price increase is the market pricing it in. The problem is what the increase does to the comparison. At $5.00, Exa was already above Brave at $5.00 and far above Parallel at $1.00. At $7.00, it is above both, and the gap to Keirolabs at $0.25 semantic and $0.10 SERP is 28x on the semantic lane and 70x on the SERP lane.
The direction of travel is the signal. Every price increase in the agent-ready tier 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 expensive side keeps raising prices, the acquirers keep paying premiums, and the teams that matter are the ones with unit economics that survive at a million queries a month.
## How does Exa pricing compare to Keirolabs, Brave, Tavily, Parallel, and Serper?
**Exa at $7.00/1K base sits above Brave at $5.00/1K and Parallel Turbo at $1.00/1K, below Tavily at $8.00/1K, and far above Keirolabs at $0.25/1K semantic and $0.10/1K SERP.** On a like-for-like basis, Exa is the second-most-expensive option in the category, and the gap to the cheap side is not small. 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, including the [Keirolabs pricing page](https://keirolabs.cloud/pricing).
| 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 |
| 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 |
| SerpAPI | $3.75-25.00 | 250 searches/mo | SERP scraping, many engines |
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. The pattern across all of them is the same: the headline price is a floor, not a bill.
The gap between the tiers is not small. At one million queries a month, the spread is the difference between a $250 invoice on Keirolabs semantic and a $7,000 invoice on Exa base search, and a $27,000 invoice on Exa's 30-result search. That is a 28x to 108x difference on the same workload, and it compounds into 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.
Monthly bill at 1K, 10K, 100K, and 1M queries. The 30-result surcharge puts Exa at $27,000 at 1M queries, 108x Keirolabs at $250.
## What is the true cost of an Exa query your agent can actually use?
**The true cost of an Exa query is search plus Contents plus extra results plus answer synthesis, and a realistic agent query lands at $0.027 to $0.05, not the $0.007 the headline implies.** Run the loop and watch the arithmetic. A search at $7.00 per 1,000 is $0.007. A Contents call for text on 10 pages is $0.01. A Contents call for highlights on the same 10 pages is another $0.01. An AI summary on top is another $0.01. The search that looked like $0.007 is now $0.037, and you have not asked a single question yet.
The answer endpoint adds more. An Answer call at $5.00 per 1,000 is $0.005, and it runs on top of the search that feeds it. A deep search at $12.00 per 1,000 is $0.012, and it runs multiple searches internally. A deep-reasoning search at $15.00 per 1,000 is $0.015, and it runs a full research loop. The endpoint prices are not alternatives to the search price. They are additions to it, because every one of them starts with a search.
| Agent query shape | Exa cost per query | Keirolabs equivalent |
|---|---|---|
| Search, 10 results | $0.007 | $0.00025 |
| Search + text contents | $0.017 | $0.00025 |
| Search + text + highlights | $0.027 | $0.00025 |
| Search + text + highlights + summary | $0.037 | $0.00025 |
| 30-result search + text | $0.037 | $0.00025 |
The same workload on the flat-rate side is one call. Keirolabs returns search results and extracted content together at $0.25 per 1,000 semantic queries, which is $0.00025 per query. Parallel returns search and excerpts at $1.00 per 1,000, which is $0.001 per query. The gap between $0.037 and $0.00025 is 148x, and it compounds into 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.
## When is Exa worth the price in 2026?
**Exa is worth its price when you need keyword-free semantic similarity retrieval, deep research, or its neural index; it is a poor value for plain web search, QA, and RAG, where Keirolabs at $0.25/1K returns the same class of result.** The honest framing is short. Exa's real differentiator is the find-similar primitive: you hand it one page and get back pages that mean the same thing even if they share zero terms. That is a specialist capability, and it is unmatched. For everything else, the price does not hold up.
| Workload | Pick | Why |
|---|---|---|
| Find-similar semantic retrieval | **Exa** $7.00/1K | Unmatched keyword-free similarity |
| Deep research, turnkey | **Exa** $12-15/1K | Multi-step research loop |
| 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 |
The waste case is specific. If you are paying Exa $7.00 per 1,000 for plain web search and then paying Contents on top to read the pages, you are paying the specialist premium for a commodity job. The teams that do this are the ones the pricing page was designed for. The teams that read the invoice are the ones who switch the commodity queries to the $0.25 lane and keep Exa for the find-similar queries nothing else can do. The hybrid setup most production teams land on is Keirolabs for QA and RAG, Exa for the occasional find-similar query, and Serper or Brave for the keyword-able majority. Splitting by query type is the biggest cost lever there is.
## How fast is the Exa Search API?
**Exa claims sub-425ms on Fast and sub-200ms on Instant, both vendor claims, which puts it in the same latency class as Parallel Turbo at ~200ms and ahead of Brave at 669ms measured.** Speed is not a footnote. In a multi-step agent loop, search latency compounds: five searches back to back is two seconds on Exa Fast and three seconds on Brave, and a slow API can turn a snappy agent into a 30-second wait.
| Provider | Median latency | How measured |
|---|---|---|
| **Exa Instant** | Sub-200ms | Vendor claim |
| **Parallel Turbo** | ~200ms | Vendor claim (p50) |
| **Exa Fast** | Sub-425ms | Vendor claim |
| Brave Search | 669ms | AIMultiple benchmark, measured |
| Tavily | 998ms | AIMultiple benchmark, measured |
The latency story is real, and it is one of the reasons Exa's price holds up for the workloads it is built for. A find-similar query on a neural index at sub-425ms is a different product from a Google SERP scrape at 669ms. The problem is that latency does not justify the price for commodity search. The vendors at the bottom of the price table are also at the top of the latency table, and the ones in the middle are measured, not claimed. When quality ties and price diverges 20x to 100x, price decides.
## 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 at signup plus $10 monthly, and Keirolabs gives you 1,000 free queries a month with no card.** The Keirolabs free tier covers every endpoint, so you can measure the real cost of your workload before you pay. The 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 and read the outputs side by side. You will see the quality difference on plain search 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 on the base search and $2,700 on a 30-result search, and Keirolabs is $25. 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 is not the price of a search, it is the price of a 10-result search, and the stack on top of it can quadruple the bill. The market split in two in 2026. You can pay $7.00 to $27.00 for a query your agent can use, or you can pay $0.25 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 the Exa Search API cost per 1,000 requests in 2026?
Exa Search API costs $7.00 per 1,000 search requests in 2026, with $5.00 per 1,000 for Answer, $12.00 per 1,000 for Deep Search, $15.00 per 1,000 for Deep-Reasoning Search and Monitors, and $1.00 per 1,000 pages on Contents.
### Does the Exa Search API have a free tier?
Yes. Exa gives $20 in credits at signup plus $10 in credits every month, roughly 1,400 basic searches a month recurring, at 5 QPS on the free tier and 10 QPS on the developer plan.
### Why is Exa more expensive than it looks?
Exa's $7.00 per 1,000 search requests covers only 10 results, and every result above 10 bills +$1.00 per 1,000, so a 30-result search costs $27.00 per 1,000. Contents bills $1.00 per 1,000 pages per content type, and the Agent API runs $0.012 to $1.00 per run.
### 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.
