Guides
Exa API pricing explained: every endpoint, the limits, and what 10,000 agent queries cost
Exa’s search API costs $4 per 1,000 requests on its lite type and $7 per 1,000 on auto, the default, with up to 10 results and their page text or highlights included. Above that you pay $1 per 1,000 extra results, $1 per 1,000 AI page summaries, $1 per 1,000 pages per content type on /contents, $5 per 1,000 on /answer, $12 to $15 per 1,000 for deep search, and $0.012 to $1.00 per run, or metered usage, on the Agent API. Every team gets $10 of free credit a month. At 10,000 agent queries a month, one lite search per query is $40 and one auto search is $70; an agent that makes three or four calls per query pays three or four times that.
Prices read November 3, 2026, logged out, on Exa’s own pricing docs, pricing page and billing and rate limits page. We make Hyperresearch, which sells a search API in the same market, so weigh the comparison at the end with that in mind. The Exa figures are Exa’s; the totals are our arithmetic from them.
Exa API prices by endpoint
Exa is pay as you go: no subscription, no minimum spend, prepaid credits drawn down per request. The base price of each request covers up to 10 results.
| Endpoint or type | Base price | Each result above 10 | AI page summaries | Typical latency |
|---|---|---|---|---|
/search, lite |
$4 per 1,000 requests | $1 per 1,000 results | Not supported | About 450 ms |
/search, auto (default) |
$7 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | About 1 s |
/search, deep-lite |
$12 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | About 4 s |
/search, deep |
$12 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | 4 to 15 s |
/search, deep-reasoning (legacy) |
$15 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | 12 to 40 s |
/search, instant (legacy) |
$4 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | |
/search, fast (legacy) |
$7 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | |
/contents |
$1 per 1,000 pages, per content type | $1 per 1,000 pages | ||
/answer |
$5 per 1,000 requests | |||
/monitors |
$2 per 1,000 requests | $1 per 1,000 results | $1 per 1,000 pages | Scheduled |
Latencies are from the search guide and the deep search table in the pricing docs. The search guide says you can request text, highlights or both “at no extra cost”; only summary costs extra, because it runs a language model on each page. numResults goes up to 100, so a 100-result auto search costs $0.007 plus 90 extra results at $0.001, or $0.097.
lite was added on October 19, 2026. It serves page content only from Exa’s cache, never fetches a page live, drops results with no cached copy when you ask for text or highlights, and returns a 400 for parameters it does not support, such as summary or maxAgeHours. Exa now recommends lite, auto and deep. The legacy types keep working at their old prices, and the changelog says lite “runs the same search as fast”, which is $3 per 1,000 cheaper.
Exa Agent API pricing
The Agent API runs asynchronous research, list building and enrichment. You choose a fixed effort for a flat price per run, or let it meter usage.
| Effort | Price per run |
|---|---|
minimal |
$0.012 |
low |
$0.025 |
medium |
$0.10 |
high |
$0.50 |
xhigh |
$1.00 |
auto (default) |
Metered, capped at $5 by default |
ultra |
Metered, capped at $20 by default |
Metered runs bill $0.10 per Agent Compute Unit, $0.005 per search tool call, $0.02 per email found and $0.07 per phone number found. budget.maxCostDollars sets the cap on auto and ultra anywhere from $1 to $100; fixed efforts take no budget. Runs that pull data through Exa Connect providers pay each provider’s fee on top, for example $0.02 per credit for Fiber.ai and $0.10 to $4.00 per order for Baselayer.
Exa’s old /research endpoint is gone: the changelog’s April 1, 2026 notice says it was replaced by /search with type: "deep-reasoning", and the October changelog points long research at the Agent API.
Free credits, plans and rate limits
The billing page lists three plans:
| Plan | Billing | Search rate limit | Agent concurrency |
|---|---|---|---|
| Free | $10 in credits every month, plus a one-time $10 onboarding bonus | 10 QPS | 50 active runs |
| Pay as you go | Prepaid credits, no subscription or minimum | 10 QPS, 25 QPS after you buy $1,000 of credits in 30 days | 50 active runs |
| Enterprise | Custom volume pricing, optional invoicing | Custom | Custom |
The free $10 resets on the first of each month and does not roll over; only a user’s first team gets it, and the onboarding bonus does not expire. $10 buys 2,500 lite searches or about 1,400 auto searches. The 25 QPS tier lasts 90 days and counts credits bought, not spent. When credits run out, or a key hits the budget an admin gave it, requests return 402 Payment Required. Exa’s pricing page also lists startup and education grants of up to $1,000 in credits.
Default rate limits apply to the whole team across all its keys:
| Endpoint | Default limit |
|---|---|
/search, /answer, /chat/completions |
10 QPS |
/search with a deep type |
5 QPS |
/contents |
100 QPS |
/agent/runs, /responses |
5 QPS and 50 active runs |
/websets/* |
20 QPS |
10 QPS is 600 searches a minute, far above an average 10,000 a month. The limit matters for bursts: a batch job or many users at once can hit it long before the monthly total looks large.
What 10,000 agent queries a month cost on Exa
Each row is one way an agent might use Exa per user query, priced at 10,000 queries a month from the rates above. These are our arithmetic, before the free $10.
| Pattern per query | Cost per query | 10,000 queries a month |
|---|---|---|
One lite search, 10 results with highlights |
$0.004 | $40 |
One auto search, 10 results with text |
$0.007 | $70 |
One auto search, 25 results |
$0.022 | $220 |
One auto search, 10 results with summaries |
$0.017 | $170 |
Three auto searches plus two /contents pages (text only) |
$0.023 | $230 |
One /answer call |
$0.005 | $50 |
One deep search |
$0.012 | $120 |
One Agent run at low effort |
$0.025 | $250 |
One Agent run at medium effort |
$0.10 | $1,000 |
One Agent run at auto, if every run hit the default $5 cap |
$5.00 | $50,000 (the ceiling, not a forecast) |
The search rows stay in the tens to low hundreds of dollars. The Agent rows are where a bill grows: a fixed medium run costs 25 lite searches, and a metered run’s cost depends on how hard the question is, which is why setting budget.maxCostDollars to the most a query is worth to you is the first thing to do on auto.
What drives an Exa bill up
- Calls per query, not queries. An agent loop that searches, reads, then searches again bills every call. Count tool calls per task in your logs before you estimate.
- The search type.
autois 75% more thanlite, and deep search is three timeslite. Legacyfastcosts $7 per 1,000 for the searchliteruns at $4. - More than 10 results. Every result past the tenth adds $0.001.
numResults: 50onautois $0.047 a call, nearly seven times the 10-result price. - Summaries.
summaryadds $0.001 a page; text and highlights add nothing on/search. - Fetching pages twice. Contents come free with the first 10 search results, and the contents guide recommends getting them that way.
/contentsbills each view separately, so text plus highlights on one page is two charges. - Metered Agent runs.
autoandultrabill compute units and searches up to their caps, and contact lookups and Connect provider calls come on top. - Retries and bursts. Past the QPS limit your client gets errors; past your credits or a key’s budget it gets
402. Retry logic that resends calls which already succeeded pays for them twice.
Exa versus other search APIs for agents
The table compares the plain search call a developer would put in an agent’s tool list, at 10 results, pay-as-you-go rates, read on November 3, 2026. The 10,000-call column is our arithmetic.
| API | Per call, 10 results | 10,000 calls | Results above 10 | Free allowance | Default rate limit |
|---|---|---|---|---|---|
| Exa | $0.004 (lite), $0.007 (auto) |
$40 to $70 | $1 per 1,000 | $10 a month, plus $10 once | 10 QPS |
| Parallel Search | $0.001 (turbo, fast), $0.005 (basic, advanced) |
$10 to $50 | $1 per 1,000 | Up to $80 at signup plus $5 a month | 600 a minute |
| Perplexity Search API | $0.001 (Fast Search), $0.005 (standard) | $10 to $50 | Up to 20 at the same price | None listed | 50 query units a second |
Hyperresearch Answers search |
5¢ (standard), 15¢ (max) |
$500 to $1,500 | Up to 25 at the same price | $5 once, no card | 60 a minute per key, 120 per workspace |
Sources: Exa pricing, Exa billing and rate limits, Parallel pricing docs, Parallel pricing page, Parallel rate limits, Perplexity pricing, Perplexity rate limits, Hyperresearch Search API docs.
On price alone, Parallel’s and Perplexity’s fast modes are the cheapest per call, Exa lite sits in between, and Hyperresearch costs 7 to 50 times more per call than the others. Perplexity bills one request for up to five queries in an array, though each query counts against the rate limit.
Hyperresearch Answers search costs more because each call does more: it searches the web and 120+ structured sources (company filings, statistics agencies, scholarly indexes, books, government records), reads the top pages live, and marks each passage cut word for word from the live page verified: true. It usually takes 5 to 10 seconds in standard mode, so it does not belong in a sub-second loop. On DeepSearchQA only, our test scored Hyperresearch search 89 vs Exa 72, with the same agent and the same grader; Hyperresearch Search was tested in standard mode as it runs in production, and the Exa row is Exa’s search API. That is one benchmark, not a general ranking. A search that fails on our side is free. If your agent wants one checked answer instead of links, the ask tool costs 2¢ (fast) to 50¢ (max) a call, against $0.005 for Exa’s /answer. Prices and install links are on the Answers page.
For a high-volume agent that reads results itself and needs sub-second latency, Exa lite, Parallel fast or Perplexity Fast Search is the cheaper fit. For fewer, harder lookups where wrong sources cost more than the call, a slower search that reads pages and structured records may cost less overall. Our comparison of nine search APIs covers MCP servers and the rest of the field.
Questions about Exa pricing
Is the Exa API free?
There is a free tier: $10 of credit every month, plus a one-time $10 bonus for finishing onboarding, with no payment method needed. Free credits are spent at the normal rates. Past that, it is pay as you go.
How much does one Exa search cost?
$0.004 on lite and $0.007 on auto, with up to 10 results and their text or highlights. Deep search is $0.012, or $0.015 on the legacy deep-reasoning type.
Does Exa charge separately for page contents?
Not for text and highlights on the first 10 search results. Summaries cost $0.001 a page, results past 10 cost $0.001 each, and the /contents endpoint charges $0.001 per page for each content type you ask for.
What happened to the Exa research endpoint?
Exa retired /research in April 2026. Its changelog points to /search with a deep type for synthesis and to the Agent API, at $0.012 to $1.00 per fixed-effort run or metered, for long research.
Does Exa offer volume discounts?
On the Enterprise plan: volume discounts, invoice billing, custom rate limits, zero data retention and up to 1,000 results per search, quoted by Exa’s sales team.
