---
title: "Exa API Pricing 2026: Official Rates per 1K, Free Tier"
category: "pricing"
url: https://keirolabs.cloud/blog/exa-api-pricing-2026
---

Exa's search API costs $7 per 1,000 requests in 2026, and the base price includes 10 results. Every result past 10 adds $1 per 1,000, so a 30-result search bills $27 per 1,000. The free tier gives you $20 in credits at signup plus $10 in credits every month, with no payment method required. This post prices every Exa endpoint, works the math at 10k, 100k, and 1M requests a month, and compares the bill against SerpAPI, Tavily, Firecrawl, Parallel, Perplexity, and Keirolabs using numbers from each vendor's own pricing page. Every rate here was checked on September 23, 2026.

## Exa Pricing

Exa sells usage, not plans. The [pricing docs](https://exa.ai/docs/admin/pricing) say it plainly: "Exa is pay-as-you-go. There is no subscription and no minimum spend: you load credits and are charged per request." You top up a dollar balance, each request draws it down at a published rate, and you can set auto-recharge so the balance refills itself. Prices are quoted per 1,000 requests, and Exa notes that a $7/1k rate works out to $0.007 per call.

Here is the whole price list on one card, straight from [Exa's pricing docs](https://exa.ai/docs/admin/pricing):

| Endpoint | Price | What it includes |
|---|---|---|
| Search | $7 / 1k requests | Real-time search with token-efficient page contents, up to 10 results |
| Deep Search | $12 / 1k requests | Multi-step research with structured outputs and citations |
| Deep-Reasoning Search | $15 / 1k requests | Same loop with more reasoning per run |
| Answer | $5 / 1k requests | An LLM answer to a question, with citations |
| Contents | $1 / 1k pages | Full page text, highlights, and summaries for known URLs |
| Monitors | $15 / 1k requests | Scheduled searches that surface new events on the web |
| Agent | $0.10 / ACU + $0.005 / search | Metered runs up to a default $20 per-run cap; fixed efforts $0.012 to $1.00 per run |

Two surcharges ride on top. The base price on search endpoints covers the first 10 results, and "every result above 10 adds $1 / 1k results" on Search, Deep Search, Deep-Reasoning Search, and Monitors. Exa-generated page summaries bill at $1 / 1k pages on any endpoint that returns one. [Exa's docs](https://exa.ai/docs/admin/pricing) confirm the Answer endpoint carries neither surcharge, which makes Answer the only flat-priced way to get a cited response.

There is also Exa Connect, the marketplace that lets an Agent pull from third-party data providers mid-run. [Exa's docs](https://exa.ai/docs/admin/pricing) list Fiber.ai at $0.02 per credit and Baselayer at $0.15 to $4.00 per order depending on the operation. Those provider calls bill separately from the run itself, so a contact-enrichment agent can spend more on Connect data than on the search that triggered it.

Pros of this model:

- No commitment. There is no monthly minimum, no seat count, and no plan ladder to climb. You can run $3 of searches and stop.
- Every rate is public and per endpoint, so you can compute a bill before you write code.
- One credit balance covers Search, Contents, Answer, Deep Search, Monitors, and Agent. No per-product invoices.

Cons:

- The headline rate is not the rate you pay the moment you ask for depth or full text. Depth 30 is $27/1k, nearly 4x the sticker.
- Contents bill per page and per content type, so one URL returned with text plus a summary bills twice.
- The meter runs on every dimension at once: requests, extra results, page types, summaries, and Connect calls. Forecasting means modeling all of them.

The honest summary: Exa does not sell you a plan; it hands you a rate card and points at the enterprise door when you outgrow it.

## Exa API Pricing 2026

The current setup has three account tiers, and the differences are about limits rather than rates. [Exa's pricing page](https://exa.ai/pricing) names them Starter (free), Developer (pay as you go), and Enterprise (custom), with 10 search queries per second (QPS) on all endpoints for self-serve accounts.

| Tier | Price | What it gates |
|---|---|---|
| Starter | Free | $20 signup credits, $10 monthly credits, 10 QPS |
| Developer | Pay as you go | Same rates, prepaid balance, auto-recharge |
| Enterprise | Custom pricing | Up to 1,000 results per search, requests above 25 results, custom QPS, custom indexes, ZDR, SLAs |

That third row is the fine print most comparisons skip. Self-serve accounts can request up to 25 results per search. [Exa's pricing docs](https://exa.ai/docs/admin/pricing) gate anything above 25 results, and searches of up to 1,000 results, behind Enterprise custom pricing. If your pipeline assumes depth 100, the calculator on the pricing page will not save you; you need a sales call.


What does not change is the billing shape. [Exa's docs](https://exa.ai/docs/admin/pricing) describe a prepaid dollar balance that usage draws down, with invoices and auto-recharge managed from the dashboard. Enterprise adds volume discounts, Zero Data Retention, custom MSA and DPA terms, and a dedicated support channel, per [Exa's pricing page](https://exa.ai/pricing). For most teams the practical question is not which tier they are on. It is whether their workload pays the surcharges, which is where the bill actually lives.

## Exa Pricing Changelog 2026

Exa's card has moved twice in 2026, and older posts still quote the old numbers. Here is what changed and when, from the [UsagePricing tracker](https://www.usagepricing.com/blueprint/exa-ai) archive of Exa's pricing pages and [Exa's current docs](https://exa.ai/docs/admin/pricing):

| When | What changed |
|---|---|
| Q2 2026 | Endpoint-card redesign. Base Search rose from $5 to $7 per 1,000 requests. Deep Search was quoted at $12 to $15 per 1,000. Answer dropped its extra-result and summary surcharges and became a flat $5/1k. |
| July 14, 2026 | Agent recut. New Minimal fixed-effort mode at $0.012 per run, X-High halved from $2.00 to $1.00, in-agent search cut to $0.005 per search, Agent Compute Units priced at $0.10/ACU. |
| Q3 2026 | No further changes to the core endpoint rates as of the September 23, 2026 check for this post. |

If you have seen Exa search quoted at $5 per 1,000, that was the pre-Q2-2026 base rate. Every per-1k figure in this post uses the current $7 card.

## Exa API Pricing Free Tier 2026

Exa runs one of the more generous recurring free tiers in the search API market, and it does not require a card. New accounts get $20 in free credits, which Exa's [pricing docs](https://exa.ai/docs/admin/pricing) estimate at around 2,800 searches, and the Free Tier adds $10 in credits every month. At the $7/1k search rate, the recurring $10 is roughly 1,428 searches a month, indefinitely, before you ever pay.

What the credits buy depends on the endpoint. The same $10 covers about 1,428 searches, about 2,000 Answers at $5/1k, or about 10,000 pages of Contents at $1/1k. A third-party [pricing tracker](https://www.usagepricing.com/tools/pricing-calculator/exa-ai) notes the monthly credits expire at month end and do not roll over, so an unused month is a lost $10. Plan your test calendar around that expiry rather than banking credits for a big evaluation later.

Rate limits are part of the value. [Exa's pricing page](https://exa.ai/pricing) advertises 10 QPS on all endpoints, which is 600 requests a minute of burst headroom, more than enough to parallelize an agent's research loop during a free evaluation.

How that stacks up against the field, from each vendor's own pricing page:

| Provider | Free tier | Card required | Recurring? |
|---|---|---|---|
| Exa | $20 signup + $10/mo credits (about 1,428 searches/mo) | No | Yes, monthly credits |
| SerpAPI | 250 searches/mo | No | Yes |
| Tavily | 1,000 credits/mo | No | Yes |
| Firecrawl | 1,000 credits/mo | No | Yes |
| Parallel | 5,000 requests/mo + $80 signup | No | Yes |
| Brave | $5 monthly credits (about 1,000 queries) | Yes | Yes |
| Keirolabs | 1,250 credits/mo = 12,500 lite searches | No | Yes |

Sources: [SerpAPI](https://serpapi.com/pricing), [Tavily](https://www.tavily.com/pricing), [Firecrawl](https://www.firecrawl.dev/pricing), [Parallel](https://parallel.ai/pricing), [Brave](https://brave.com/search/api/), [Keirolabs](https://keirolabs.cloud/pricing).

Exa's free tier wins on flexibility because credits spend on any endpoint at the same dollar value. Keirolabs' free tier wins on raw search volume: 12,500 lite searches a month is about 9x Exa's recurring search equivalent, though Keiro's free rate limit is 30 requests a minute against Exa's 10 QPS, so bursty evaluation workloads feel the difference. Brave is the odd one out for requiring a card before you spend its credit.

Building on the free tier is viable for prototypes and cron jobs, with two cautions. First, monthly credits expire, so a quiet month is a wasted month. Second, the free tier runs on the same price list as paid usage, which means the surcharges apply during evaluation too. If you test with depth 30, you are testing a product you have not priced yet.

## Exa API Pricing Neural Search Credits

Exa started as a neural search engine: an index ranked by embeddings rather than by keywords, built to answer queries like "pages similar to this blog post" or "technical posts about vector databases published this month." That product history is why people search for Exa's "neural search credits" today. The current [pricing docs](https://exa.ai/docs/admin/pricing) treat neural retrieval as the default search mode, billed at the same $7/1k request rate as everything else, with no separate credit tier for neural versus keyword queries.

The credit math has three layers, and the second layer is where budgets break.

Layer one is the request. Every endpoint has a base price per request that includes up to 10 results, per [Exa's docs](https://exa.ai/docs/admin/pricing).

Layer two is the result surcharge. Results 11 and beyond bill at $1/1k each. A depth-10 search costs $7/1k. Depth 20 costs $17/1k. Depth 30 costs $27/1k. Depth 100 costs $97/1k. The eleventh result is where the bill starts to move.

Layer three is content. Full page contents bill at $1/1k pages, and Exa bills each content type separately, so a page returned with full text plus an Exa-generated summary bills twice: once for the text and once for the summary at $1/1k pages on any endpoint that returns one ([pricing docs](https://exa.ai/docs/admin/pricing)).

A worked example for a typical [RAG pipeline](/blog/best-search-apis-for-rag). Your agent searches at depth 10 ($7/1k) and pulls full text for all 10 results ($10/1k pages at $1/1k per page). That is $17/1k requests, or $0.017 per search-with-contents call. Add a summary to each page and the contents layer becomes $20/1k, for $27/1k all in. The same call at depth 30 with text and summaries runs $27 for search plus $60 for contents: $87/1k requests. The search was never the expensive part.

For teams that want the neural retrieval without managing any of this per-dimension math, the Answer endpoint trades control for simplicity: one question in, one cited LLM answer out, flat $5/1k with no result or summary surcharges, per [Exa's pricing docs](https://exa.ai/docs/admin/pricing). It returns citations rather than ranked pages, so it fits question-answering surfaces better than ingestion pipelines.

## Exa Search Pricing

The Search endpoint is the product most people buy, so it deserves its own math. The base rate is $7/1k requests, and Exa describes what that buys as "real-time search with token-efficient page contents" for up to 10 results ([pricing docs](https://exa.ai/docs/admin/pricing)). The token-efficient part matters for agents: each result carries a compressed excerpt designed to feed an LLM without a separate fetch.

Depth is the first lever, and it is priced linearly past the base:

| Results per search | Price per 1k requests | Multiplier vs base |
|---|---|---|
| 10 | $7.00 | 1.0x |
| 20 | $17.00 | 2.4x |
| 25 | $22.00 | 3.1x |
| 30 | $27.00 | 3.9x |
| 50 | $47.00 | 6.7x |
| 100 | $97.00 | 13.9x |

Beyond 25 results you hit the Enterprise gate, since requests above 25 results require custom pricing per [Exa's docs](https://exa.ai/docs/admin/pricing). The rows past 30 are the arithmetic of the $1/1k surcharge, useful for budgeting an Enterprise quote against.

Flip the same numbers around and you get cost per result, which is the steadier metric. At depth 10 you pay $0.70 per 1,000 results. At depth 30 you pay $0.90. At depth 100 you pay $0.97. Exa's price per result barely moves with depth; its price per request nearly triples by depth 30 and multiplies fourteen-fold by depth 100. If you compare vendors on price per request, you are comparing the wrong axis. Compare on price per usable result after deduplication and dead-link filtering.

Contents and summaries are the second lever, and they attach to any endpoint. Contents at $1/1k pages covers full page text, highlights, and summaries for known URLs, and page summaries bill at $1/1k pages on any endpoint that returns one ([pricing docs](https://exa.ai/docs/admin/pricing)). A search-plus-contents pipeline therefore has two line items where a bare search has one, and the contents line is usually the bigger of the two.

One structural note in Exa's favor: the search is real-time with live retrieval behind the neural index, so results reflect pages published today without a separate crawl step. The trade is that you cannot buy bulk depth on self-serve. If your workload is "give me 100 fresh results per query, all day," Exa's pricing page routes you to Enterprise, and competitors like [Brave](https://brave.com/search/api/) or [Keirolabs](https://keirolabs.cloud/pricing) sell that depth on standard plans.

## Exa API Cost

Here is the bill at three volumes, for the configurations teams actually run. All figures come from the [published rates](https://exa.ai/docs/admin/pricing): search $7/1k plus $1/1k per result past 10, contents $1/1k pages, answer $5/1k, Deep Search $12/1k, Deep-Reasoning $15/1k, Monitors $15/1k.

| Configuration | Per 1k requests | 10k / mo | 100k / mo | 1M / mo |
|---|---|---|---|---|
| Search, depth 10 | $7.00 | $70 | $700 | $7,000 |
| Search, depth 30 | $27.00 | $270 | $2,700 | $27,000 |
| Search depth 10 + page text | $17.00 | $170 | $1,700 | $17,000 |
| Search depth 10 + text + summaries | $27.00 | $270 | $2,700 | $27,000 |
| Answer | $5.00 | $50 | $500 | $5,000 |
| Deep Search | $12.00 | $120 | $1,200 | $12,000 |
| Deep-Reasoning Search | $15.00 | $150 | $1,500 | $15,000 |
| Monitors | $15.00 | $150 | $1,500 | $15,000 |

The same volumes on Keirolabs' /search/lite, for reference: the list rate is $0.25/1k ([pricing](https://keirolabs.cloud/pricing)), so 10k requests cost $2.50, 100k cost $25, and 1M cost $250 on the headline. On monthly plans lite bills at 0.1 credit per request: $0.24/1k on Essential ($30/mo, 125,000 lite searches), $0.13/1k on Pro ($50/mo, 375,000), and $0.08/1k on Startup ($100/mo, 1,250,000). A million lite searches a month fits inside the $100 Startup plan with room to spare. Exa's depth-10 bill for the same volume is $7,000, and its depth-30 bill is $27,000. That gap is the whole reason pricing-comparison content about Exa exists.

Three levers cut an Exa bill without changing vendors.

First, cap numResults at 10 unless a workload provably needs more. Dropping from depth 30 to depth 10 on a 100k-request pipeline saves $2,000 a month, which is a 74% cut, arithmetic you can verify against the [rate card](https://exa.ai/docs/admin/pricing).

Second, question whether each pipeline stage needs summaries. The $1/1k summary surcharge is cheap per page and merciless at volume: 1M pages summarized is $1,000.

Third, replace polling loops with Monitors. An agent that re-runs the same search hourly burns $15/1k in Monitors instead of stacking search-plus-contents calls, and Exa built the endpoint for exactly that trade.

The scale test that matters: at 10k requests a month, Exa's depth-10 bill is a rounding error and the API is a joy. At a million requests a month, the surcharge is the invoice, and the per-line arithmetic above is what your finance team will actually read.

## Exa vs SerpAPI

SerpAPI and Exa sell different products that happen to both return links. SerpAPI delivers structured Google SERP data: organic results with positions, snippets, knowledge graph panels, and Google verticals like Maps, News, and Jobs. Exa delivers neural-ranked pages from its own index. The price difference is a subscription model against a meter, and each hides its own trap.

SerpAPI's plans, from [its pricing page](https://serpapi.com/pricing): Free at 250 searches a month, Starter at $25/mo for 1,000 searches, Developer at $75/mo for 5,000 searches, continuing up the ladder to the largest self-serve plans around $275/mo for 30,000 searches. That is $25/1k at the bottom of the ladder, $15/1k on Developer, and about $9.17/1k at the top.

| | Exa | SerpAPI |
|---|---|---|
| Entry paid rate | $7/1k requests | $25/1k searches (Starter) |
| Best self-serve rate | $7/1k (volume discounts via Enterprise) | about $9.17/1k at 30k/mo |
| Result count billing | +$1/1k per result past 10 | Result count does not change the price |
| 100-result search | $97/1k | 1 search credit |
| Free tier | $20 signup + $10/mo credits | 250 searches/mo |
| Index | Own neural index | Google's SERP, scraped |
| Caching | No cached-search discount | Cached searches do not burn credits |

The result-count line is the one to read twice. [SerpAPI's pricing page](https://serpapi.com/pricing) states that responses with 100 results or empty result sets both count as 1 search. Exa's 100-result search bills $97/1k. A depth-100 workload costs 13x more on Exa's meter than on SerpAPI's subscription, before Enterprise pricing enters the conversation. SerpAPI also does not bill cached searches, so repeated identical queries are effectively free, a pattern that matters for monitoring workloads.

SerpAPI pros:

- One credit per search no matter how many results come back, which makes depth cheap.
- Google's ranking and freshness, including Maps, News, Jobs, and Trends endpoints.
- Cached searches are free, which rewards polling workloads.
- 24/7 support, Legal Shield options, and ZeroTrace mode on higher plans ([pricing](https://serpapi.com/pricing)).

SerpAPI cons:

- The entry rate is $25/1k, more than 3x Exa's sticker, and $15/1k at the tier most teams land on.
- It is a subscription, so a quiet month costs the same as a busy one.
- The input is scraped Google, which carries legal exposure Google has actively litigated; [our crackdown post](/blogs/guide/search-api-crackdown-2026-serpapi-lawsuit-bing-shutdown) tracks the case.
- No semantic recall. It finds pages matching keywords, not meaning.

Exa pros:

- Neural ranking finds pages that keyword queries miss, and find_similar-style lookups are a query type SerpAPI cannot answer at all.
- Pay-as-you-go with no minimum, so costs track usage to zero.
- Contents, answers, and deep research run on the same balance.

Exa cons:

- The per-result surcharge punishes exactly the deep-coverage workloads SERPs are good at.
- No cached-search discount, so polling the same query bills full price every time.
- Depth past 25 results is an Enterprise conversation.

Migration between them is a two-day job, not a rewrite. Moving from SerpAPI to Exa, expect to translate query strings into natural language plus filters (includeDomains, date ranges, category pins), map organic_results fields to Exa's results array, and accept that position ordering reflects neural relevance rather than Google's rank. Moving from Exa to SerpAPI, port your depth: a SerpAPI num=100 request replaces what would be a $97/1k Exa call, and your bill drops. Teams running both usually land on a split: SerpAPI for rank-tracking and SERP features, Exa for semantic retrieval inside agent loops.

## Tavily vs Exa vs Firecrawl

These three are the default RAG trio, and they price the same job three different ways. Tavily sells credits, Firecrawl sells credits with a crawl engine behind them, and Exa sells requests with surcharges. The table prices each from the vendor's own pages, with SimpleQA scores from the third-party and self-published runs collected on [our deep search benchmark](/Deep-search).

| | Exa | Tavily | Firecrawl |
|---|---|---|---|
| Basic search | $7/1k, 10 results in base ([pricing](https://exa.ai/docs/admin/pricing)) | $8/1k basic, $16/1k advanced ([pricing](https://www.tavily.com/pricing)) | 2 credits per 10 results ([pricing](https://www.firecrawl.dev/pricing)) |
| Effective PAYG rate | $7.00/1k | $8.00/1k (credit = $0.008) | $10/1k PAYG ($5 per 1,000 credits) |
| Cheapest plan rate | $7/1k | $7.50/1k (Project tier) | $6.40/1k (Hobby, $16/mo for 5,000 credits) |
| Free tier | $20 + $10/mo credits | 1,000 credits/mo | 1,000 credits/mo |
| Output | Ranked pages, contents, answers, deep research | Chunks tuned for RAG, answers | Search plus full-page markdown |
| SimpleQA | 91.9% (third-party agent run) | 93.3% (self-published) | 94.7% (third-party agent run) |

Sources for the benchmark rows: the [Van Zyl third-party run](https://jayvanzyl.me/i-tested-firecrawl-exa-parallel-and-claude-search-on-simpleqa-heres-what-scored-best/) covering Firecrawl and Exa, and [Tavily's own evaluation](https://www.tavily.com/blog/tavily-evaluation-part-1-tavily-achieves-sota-on-simpleqa-benchmark). Note the harnesses differ; treat the column as directional, not decisive.

Firecrawl's search pricing deserves its own sentence, because the credits-to-searches conversion is where people get surprised. Search costs 2 credits per 10 results, and pay-as-you-go top-ups run $5 per 1,000 credits, which makes PAYG search $10 per 1,000 searches ([Firecrawl pricing](https://www.firecrawl.dev/pricing)). The Hobby plan's 5,000 credits at $16/mo (annual billing) works out to $6.40/1k. Firecrawl's real pitch is that every result arrives with full-page markdown already fetched, a job Exa would bill as search plus contents ($17/1k at depth 10). On that comparison Firecrawl's $6.40 to $10 range beats Exa's bundle.

Tavily pros:

- Chunks return RAG-ready, sized for context windows, no second extraction call.
- 1,000 recurring credits a month, no card, and a 180 ms p50 claim on /search that, if it holds on your queries, is the fastest in the group.
- Advanced search at $16/1k for when basic quality falls short.

Tavily cons:

- Advanced search doubles the rate, and most teams end up on advanced after a month of basic.
- No published neural-index story; it is a hybrid of crawl, cache, and ranking.
- The 93.3% SimpleQA figure is self-published, on a different harness than the third-party runs.

Firecrawl pros:

- Search and extraction in one call, one credit balance, full markdown out.
- Cheapest plan-rate search of the three at $6.40/1k.
- 94.7% on a third-party SimpleQA run, the best score in this trio.

Firecrawl cons:

- Live crawling means fresh pages but slower, spikier latency than a cached index.
- The credit-to-search conversion (2 credits per 10 results) is easy to misbudget; PAYG is $10/1k, double the plan rate.
- Less depth control than Exa for similarity-style queries.

Exa pros:

- The strongest semantic retrieval and find-similar capability of the three.
- Deep Search and Answer give you researched, cited outputs rather than raw results.
- The free tier's dollar value beats both rivals' credit allowances.

Exa cons:

- The most complex bill of the three: requests, extra results, content types, and summaries all meter separately.
- Depth past 25 results needs Enterprise.
- Its SimpleQA score trails both rivals on the runs collected above.

Pick by shape of work. Extraction-heavy pipelines that live on page text: Firecrawl. RAG chunks with latency ceilings: Tavily. Semantic recall, similarity search, and research loops: Exa. The full field is priced per 1k in [the Exa alternatives comparison](/blog/exa-alternatives-2026).

## Exa vs Tavily

Head to head, the sticker prices are $7 versus $8 per 1,000 basic searches, from [Exa](https://exa.ai/docs/admin/pricing) and [Tavily](https://www.tavily.com/pricing) respectively. That 14% gap is the least interesting line on either page. The differences that move invoices sit underneath it.

Depth first. Tavily's basic search returns a fixed result set per credit, and advanced search doubles the spend to $16/1k. Exa's depth scales linearly: $17/1k at 20 results, $27/1k at 30. A pipeline pulling 20 results pays Tavily $8 against Exa's $17, assuming Tavily's basic result count covers the need. A pipeline pulling 10 pays Exa $7 against Tavily's $8. The crossover sits almost exactly where agents actually operate, so run your own depth histogram before assuming either way.

Latency second. Tavily publishes a 180 ms p50 on /search and markets itself on speed; Exa publishes no latency figure at all. If your product puts search inside a user-visible loop, that published number is worth testing on your own queries before you commit, because Exa's neural retrieval tends to add a round of ranking work that a cached hybrid can skip.

Free tier third, and it is closer than it looks. Exa's recurring $10 in credits is worth about 1,428 searches at the $7 rate. Tavily's 1,000 credits are worth 1,000 basic searches. Exa's allowance also spends on Answer, Contents, and Deep Search, which makes it the better evaluation budget for teams still deciding what they need. The trade: Tavily's credits convert to RAG-shaped chunks out of the box, while Exa's search results need the contents surcharge to become ingestion-ready.

Quality is the column with an asterisk. On SimpleQA, Tavily self-publishes 93.3% using GPT-4.1 answers on the full question set, while Exa's 91.9% comes from a third-party agent run collected on [our benchmark page](/Deep-search). Different harnesses, different answering models, different question samples. The honest read: both are strong, neither number was produced under identical conditions, and the tiebreaker should be your own eval set.

When Exa wins: meaning-shaped queries, find-similar retrieval, research loops that benefit from Deep Search's citation-backed synthesis, and workloads that stay at depth 10. When Tavily wins: latency-sensitive RAG, teams that want chunked output without a contents surcharge, and pipelines that lean on advanced search quality more than on neural recall. We keep a longer breakdown in [the Tavily alternatives post](/blog/tavily-alternatives-2026), and the cheaper-than-Tavily field is surveyed in [this comparison](/blog/search-apis-cheaper-than-tavily).

## Exa vs Perplexity

Perplexity's API answers questions; Exa's API returns pages. The pricing reflects that. Perplexity's Search API bills $5 per 1,000 requests with model tokens billed separately, per its [API docs](https://docs.perplexity.ai), and the docs state the formula directly: total cost per query equals token costs plus a request fee that varies by search context size. Exa's Answer endpoint, the closest functional match, is a flat $5/1k with no token metering, since Exa runs the synthesis on its own side.

The comparison is complicated by a calendar. Perplexity's docs say the Sonar Chat Completions models are supported until September 27, 2026, with the Agent API positioned as the successor. If you are evaluating Perplexity search this quarter, you are evaluating a platform mid-migration, and the request-fee structure of the successor, not the legacy rate card, is what your invoice will follow. Budget accordingly and read the [migration guidance](https://docs.perplexity.ai) before signing anything annual.

On measured answer quality, the runs collected on [our SimpleQA benchmark](/Deep-search) score Perplexity at 85.9% against Exa's 91.9% from the third-party agent harness, with Perplexity's figures coming from the GPT-4.1 answering pipeline published in [Tavily's eval repo](https://github.com/tavily-ai/tavily-search-evals). A 6-point gap on short-form factuality is meaningful for agents that answer from search without their own reasoning step.

Perplexity pros:

- The answer arrives synthesized with citations, so your pipeline writes no synthesis code.
- $5/1k request fee is the lowest quoted request rate among answer APIs.
- The Router API bills purely per token with no request fees, a clean model for mixed-model workloads.

Perplexity cons:

- Token costs ride on top of the request fee, and long answers move the bill in ways a per-request quote hides.
- The search context size you choose changes the request fee, so cost forecasting needs a usage histogram, not a rate card.
- Sonar's deprecation deadline lands September 27, 2026.

Exa pros:

- Flat $5/1k Answer with no token surcharge, since inference happens on Exa's side.
- You can bypass synthesis entirely and take ranked pages at $7/1k when your own LLM does the reading.
- Contents and Deep Search integrate into the same balance and SDK.

Exa cons:

- Answer returns one cited response, not Perplexity's full synthesized report format.
- Deep-Reasoning at $15/1k is the only way to get heavier research behavior, and it triples the Answer rate.

The practical split: if your product is a chat surface that needs a finished answer with citations, Perplexity's request-plus-tokens model is purpose-built, migration deadline and all. If your agents need pages, excerpts, and control over synthesis, Exa's meter is the better fit, and you can read the full field in [the top-8 comparison](/blogs/comparisons/top-8-ai-search-apis-compared-2026).

## Exa Alternatives Parallel

Parallel is the closest structural rival to Exa's Search endpoint, because it also prices per request instead of per subscription, and it targets the same agent-loop workload. The rates, from [Parallel's pricing page](https://parallel.ai/pricing): Search API requests run $0.001 to $0.005 each, which is $1 to $5 per 1,000 for 10 results, with published latency bands from 200 ms to 3 s depending on the speed tier you select. Deep-research tiers run from $0.005 to $2.40 per request. The free tier is 5,000 requests a month plus an $80 signup credit and $5 in monthly credits, per [Parallel's pricing](https://parallel.ai/pricing).

Set against Exa, the shape of the deal inverts. Exa's $7/1k base includes 10 results and bills surcharges past that; Parallel's $1 to $5/1k range scales with query difficulty rather than result count, and its base request covers 10 results the same way. For a depth-10 workload, Parallel's floor is 7x cheaper than Exa's sticker. For a hard query at the top of Parallel's difficulty pricing, the two converge near $5 versus $7, and Exa's contents and Answer endpoints start to look economical again.

The benchmark record is the interesting part. On the third-party agent runs collected on [our SimpleQA benchmark](/Deep-search), Parallel scores 91.0% at an effective $1.00 per 1,000 requests, against Exa's 91.9% at $12/1k in the same harness family. A 0.9-point quality gap at a 12x price difference is the kind of number that reshorts procurement decks. Our own [head-to-head receipts](https://keirolabs.cloud/bench/keiro-lite-vs-parallel-turbo) cover the Parallel turbo tier separately, where /search/lite won 91 of 100 queries at a fraction of the cost.

Parallel pros:

- Per-request pricing with latency bands you can select, so cost and speed are explicit engineering choices.
- The cheapest credible agent-native search floor in the market at $1/1k.
- The largest free evaluation budget in the category: 5,000 requests a month plus $80 in signup credit ([pricing](https://parallel.ai/pricing)).

Parallel cons:

- The $1 to $5 range means the sticker is a floor, and difficult-query workloads drift toward the ceiling.
- No neural find-similar capability; it is a different retrieval model than Exa's embeddings.
- Deep research tiers reaching $2.40 per request need their own budget line.

Exa pros:

- Semantic recall and similarity search that Parallel's model does not attempt.
- Contents, Answer, and Deep Search on one balance, so a research pipeline needs one vendor.
- The 25-result self-serve ceiling beats nothing, but Parallel's per-request simplicity beats it for shallow loops.

Exa cons:

- At depth 10 and under, Exa's $7/1k is 2x to 7x Parallel's effective rate for the same result count.
- No published latency figure to hold against Parallel's 200 ms band.

The pick is workload-shaped. Fast, shallow, high-volume agent lookups: Parallel, starting at $1/1k. Meaning-based retrieval, similarity search, and research with citations: Exa, and pay the $7 knowingly. If neither price works, [the full alternatives field](/blog/exa-alternatives-2026) prices eight more vendors per 1k, and [the Parallel pricing deep dive](/blog/parallel-findall-api-pricing) covers the FindAll side of Parallel's catalog.

## Exa Alternatives

Every alternative below prices per 1,000 basic search requests, from each vendor's own pricing page, checked September 23, 2026.

| Provider | $/1k basic search | Free tier | What it is |
|---|---|---|---|
| Keirolabs /search/lite | $0.25 list; $0.24/1k on Essential, $0.13/1k Pro, $0.08/1k Startup | 1,250 credits/mo = 12,500 lite searches, no card | Own index, agent-ready results |
| Serper | $1.00, down to $0.30 at volume | 2,500 queries | Scraped Google SERP |
| Parallel | $1.00 to $5.00 | 5,000 req/mo + $80 signup | Per-request agent search |
| Brave | $5.00 | $5 monthly credits, card required | Independent index, 30B+ pages |
| Exa | $7.00 (10 results) | $20 + $10/mo credits | Neural index |
| Firecrawl | $6.40 plan to $10 PAYG | 1,000 credits/mo | Search + full-page markdown |
| Tavily | $8.00 basic / $16.00 advanced | 1,000 credits/mo | RAG-shaped hybrid |
| SerpAPI | $15.00 (Developer) | 250 searches/mo | Structured Google SERP |

Sources: [Keirolabs](https://keirolabs.cloud/pricing), [Serper](https://coldiq.com/blog/serper-pricing), [Parallel](https://parallel.ai/pricing), [Brave](https://brave.com/search/api/), [Exa](https://exa.ai/docs/admin/pricing), [Firecrawl](https://www.firecrawl.dev/pricing), [Tavily](https://www.tavily.com/pricing), [SerpAPI](https://serpapi.com/pricing).

Keirolabs is ours, so here are the numbers with the catches attached. /search/lite has a $0.25/1k list rate, and monthly plans bill it at 0.1 credit per request: $0.24/1k on Essential ($30/mo for 125,000 lite searches), $0.13/1k on Pro ($50/mo for 375,000), and $0.08/1k on Startup ($100/mo for 1,250,000), per [our pricing page](https://keirolabs.cloud/pricing). The catch: one-time credit packs bill lite at 0.5 credit per request, which lands at $2.50 to $3.33/1k, so the monthly plan is the cheap path and pack buyers pay more per search than the headline suggests. /search/fast runs 1 credit ($2.40/1k on Essential, $0.80/1k on Startup), /search/content runs 3 credits ($7.20/1k on Essential, $2.40/1k on Startup), and answers with citations run 5 credits ($12/1k on Essential, $4/1k on Startup). One honest loss: Exa's Contents endpoint at $1/1k pages undercuts our extract pricing at scale, and if your pipeline is page-content-shaped rather than search-shaped, that Exa line item is the better rate.

On measured quality, our deep search scored 95.3% on the 1,000-question seeded SimpleQA run, retrieval-only with no reader model, against Exa's 91.9% from the third-party agent harness, all collected on [the benchmark page](/Deep-search). That is our own harness and you should weight it accordingly, but the run script is open and the leaderboard is public.

Serper, at $1/1k with 2,500 free queries, is the price floor for raw Google results, and it comes with the scraped-SERP dependency documented in [the crackdown post](/blogs/guide/search-api-crackdown-2026-serpapi-lawsuit-bing-shutdown). Brave at $5/1k is the independent-index pick with the card-on-file requirement and the post-February-2026 free tier detailed in [our Brave pricing post](/blogs/comparisons/brave-search-api-pricing-2026). The full field, with latency claims and fine print, is in [the best Exa alternatives post](/blog/exa-alternatives-2026).

The boring takeaway: Exa's pricing is honest about its rates and complicated about its dimensions. Request price, result surcharge, content types, summaries, and Connect calls all meter separately, and the invoice is the product of all five. Teams at 10k requests a month should pick Exa for the neural index and not think about the bill. Teams at scale should model their depth and contents mix against the [rate card](https://exa.ai/docs/admin/pricing) before the month ends, then check the same math against Keirolabs, Parallel, or Serper, because at volume the difference is not a discount. It is the budget. Every number in this post links to its source and was checked on September 23, 2026; prices change, so check them again before you sign anything, including with us.