Both are pay-as-you-go web APIs for AI agents on their own index. Pick Parallel for the lowest search price ($1 per 1,000 in fast mode), research runs that return fields with citations, and 600 searches a minute. Pick Exa for company and people search modes, a flat-priced async research Agent and $10 of free credit every month.
Pick Exa if
Pick Parallel AI if
Exa
Research, Scraping
Parallel AI
Research, Scraping
Exa
Parallel AI
15 points side by side. Open a value for the full detail.
| What | Exa | Parallel AI |
|---|---|---|
| Web search API on its own indexA web search API on its own index. One key covers search with page contents, contents for URLs you already have, cited answers, scheduled monitors, an async research Agent and Websets for entity lists. | Web APIs for AI agents, own indexWeb APIs for AI agents on Parallel's own index: Search, Extract, Task for research and structured enrichment with citations, Responses, Monitor, FindAll for entity lists and Entity Search. | |
| Pay as you go, prepaid balancePay as you go from a prepaid dollar balance. No subscription and no minimum. Verified 2026-10-25. | Pay as you go, prepaid balancePay as you go from a prepaid dollar balance, priced per request, with optional auto-reload. Monthly spend limits only notify; they do not block requests. Verified 2026-10-25. | |
| Lite $4, auto $7, deep $12Lite $4, auto $7, deep $12, deep-reasoning $15. The price covers 10 results; each extra result costs $1 per 1,000. Verified 2026-10-25. | Turbo and fast $1, basic $5Turbo and fast $1, basic and advanced $5. The price covers 10 results; each extra result costs $1 per 1,000. The API default is advanced, so set the mode on every call. Verified 2026-10-25. | |
| $1 per 1,000 pages per type/contents costs $1 per 1,000 pages per content type, so text plus highlights on one page counts as 2 units. | $1 per 1,000 URLsExtract costs $1 per 1,000 URLs and takes up to 20 URLs per call. | |
| Agent runs from $0.012 flatExa Agent is async. A fixed effort gives a flat price per run: minimal $0.012, low $0.025, medium $0.10, high $0.50, xhigh $1.00. The auto and ultra modes are metered with default caps of $5 and $20 per run. /answer returns one cited answer for $5 per 1,000. | Task runs from $5 per 1,000Task runs return the fields you define, each with citations, reasoning and confidence, charged once per successful run: from $5 per 1,000 on lite to $2,400 per 1,000 on ultra8x. Responses gives cited answers for $10 to $250 per 1,000. | |
| Company and people search categoriesSearch has category=company and category=people modes; Exa says people search spans 1B+ public profiles. Websets builds entity lists on a separate paid Websets plan whose prices show only after login. | FindAll from $0.25 per run, betaFindAll builds verified entity lists from a plain-language description: $0.25 per run plus $0.03 per match on base, up to $10 plus $1 per match on pro. Entity Search costs $5 per 1,000 requests with 100 results each. Both are in public beta. | |
| $10/mo plus $10 bonus; card terms varyThe docs list $10 of credits a month plus a one-time $10 onboarding bonus with no card. In the app on 25 Oct 2026 a new account showed a $0.00 balance, with the free credits pending until a non-prepaid card is added. | $80 signup credit, no cardA new organization on 25 Oct 2026 received an $80 signup credit at once with no card, expiring about 60 days later. Adding a card unlocks $5 of credit a month. Startups can apply for up to $250. | |
| 10 queries/sec on searchSet per team across all keys: 10 queries a second on search and answer, 5 on deep search, 100 on /contents, 50 concurrent Agent runs. Buying $1,000 of credits within 30 days raises search to 25 a second for 90 days. | Search and Extract 600/minPer product, in a rolling 60-second window, counting only POSTs that create work: Search and Extract 600 a minute, Task runs 2,000 a minute, Monitor creation 300 a minute. Responses carry no rate-limit headers. | |
YesOfficial hosted server at https://mcp.exa.ai/mcp. Runs keyless, with OAuth or with an API key. Tools: web_search_exa, web_fetch_exa, web_search_advanced_exa and agent_run (agent_run needs a key or OAuth). | YesSearch MCP at https://search.parallel.ai/mcp is free with no account and offers web_search and web_fetch. A separate Task MCP at https://task-mcp.parallel.ai/mcp needs a key or OAuth and bills at Task prices. | |
| Scheduled monitors, $15 per 1,000Monitors run searches on a schedule and push new results to a webhook, at $15 per 1,000 requests. | Monitor from $3 per 1,000 checksMonitor checks the web on a schedule and sends events to a webhook, at $3 per 1,000 checks on lite and $10 on base. Each execution counts as one check. | |
| Webhooks; Batch API in betaMonitors and Websets deliver to webhooks. A Batch API in beta runs many searches or Agent runs and returns JSONL. | Signed webhooks; Task Groups up to 1,000Webhooks for Task runs, FindAll and Monitor, signed under the Standard Webhooks format. Task Groups add up to 1,000 runs per request, with the balance checked for the whole batch up front. | |
| costDollars field on every responseEvery response carries a costDollars field. The dashboard says recent usage can take a few hours to show. | Usage CSV shows exact costThe in-app balance rounds to the cent; the usage CSV export shows the exact cost per call. A plain API key cannot read the balance endpoint, which needs an OAuth token. | |
| Zero retention and HIPAA on EnterpriseZero data retention and HIPAA are Enterprise options. | No caching or resale of outputsOutputs are primarily for the user who ran the query, may not be cached, stored or shared with third parties for AI and data-selling services, and may not be used to create databases or data-selling products. On standard terms Parallel may train on de-identified inputs and outputs; enterprise zero data retention is excluded. | |
| SOC 2 Type IISOC 2 Type II, per its security overview. | SOC 2 Type 1 and Type IISOC 2 Type 1 and Type II certified as of April 2025, per its FAQ. EU data residency is available on enterprise agreements. | |
| Company and people search, fixed-price runsAgents that search for companies and people, need fixed-price research runs, and want free monthly credit to start. | High-volume search, cited enrichment fieldsHigh-volume agent search at the lowest unit price, and row-by-row research or enrichment that returns cited fields. |
No row matches in this view.
US dollars as each vendor's page showed them on 25 Oct 2026. One panel per tool; switch each between monthly and annual billing (an annual price is the monthly rate when you pay for the year up front).
$0
Free
$10 of credits a month plus a one-time $10 onboarding bonus with no payment method required, per the docs.
In the app on 25 Oct 2026 a new account showed a $0.00 balance, with the free credits pending until a non-prepaid card is added.
Usage-based
Prepaid dollar balance, billed per request.
Search lite $4, auto $7, deep $12, deep-reasoning $15 per 1,000; contents $1 per 1,000 pages per content type; answer $5 per 1,000; monitors $15 per 1,000; Agent from $0.012 per run.
Custom quote
Custom volume pricing, postpaid invoicing, zero data retention.
$0
Free
$80 signup credit at once with no card, expiring about 60 days after signup (in the app on 25 Oct 2026); $5 a month once a card is added.
Requests stop when credit runs out.
Usage-based
Prepaid dollar balance, billed per request.
Search $1 (turbo, fast) or $5 (basic, advanced) per 1,000; Extract $1 per 1,000 URLs; Task $5 to $2,400 per 1,000 runs; Monitor $3 to $10 per 1,000 checks; FindAll from $0.25 per run plus $0.03 per match.
Up to $250 in free credits by application.
Custom quote
Custom pricing, zero data retention, DPA, SSO, custom rate limits, EU data residency.
If neither fits: Also consider Tavily if you need site map and crawl in the same API, or a monthly credit plan with a free tier of 1,000 credits.
Choosing between Exa and Parallel AI, or moving from one to the other? Tell us your setup. We will show you how we would run it, and what we would check first.
In its fast and turbo modes, Parallel Search costs $1 per 1,000 requests with 10 results; its basic and advanced modes cost $5, and advanced is the API default. Exa's cheapest search, lite, costs $4 per 1,000. On both, each result past the first 10 adds $1 per 1,000.
Both have a keyless MCP server. Parallel's Search MCP at https://search.parallel.ai/mcp is free and anonymous with lower limits. Exa's MCP at https://mcp.exa.ai/mcp also runs keyless, rate-limited, and its research tool needs a key or OAuth.
Parallel's Task API is built for it: you define the fields, and each run returns them with citations and confidence, from $5 per 1,000 runs on lite, with up to 1,000 runs per Task Group request. Exa's Agent also does list building and enrichment, at a flat price per run from $0.012.
Check the terms first. Parallel's customer terms say outputs are primarily for the user who ran the query, may not be cached, stored or shared with third parties for AI and data-selling services, and may not be used to create databases. Exa offers zero data retention as an Enterprise option.
Parallel gave a new organization $80 at once with no card on 25 Oct 2026, expiring about 60 days later. Exa's docs list $10 a month plus a one-time $10 bonus, but a new account on the same day showed $0.00 until a card was added.
Read from exa.ai/docs/admin/pricing, docs.parallel.ai/getting-started/pricing and each tool's API docs. Verified 25 Oct 2026.
