Exa vs Tavily: Search and Extract APIs for AI Agents
Compare Exa and Tavily web search APIs for AI agents. Learn differences in search types, page extraction, credit billing, and how to choose the right tool.
Exa vs Tavily: Search and Extract APIs for AI Agents
If you are comparing Exa vs Tavily, you are usually choosing a web search API for AI agents — not a consumer search engine, and not a general web crawler. Both officially sell real-time search that returns URLs plus page content your model can use. The fork is how they bill, how they package “search vs extract vs research,” and which extras (site crawl, monitors, people/company enrichment) you actually need.
This page compares Exa and Tavily as a head-to-head. It is written for people wiring search into agents, RAG, or research workflows who care about official pricing, rate limits, and documented endpoints. It does not rank search quality. Both vendors publish their own benchmarks; we did not reproduce them.
Last checked: 2026-11-10 (Asia/Shanghai), against official product, pricing, and docs pages listed at the end. This page is an official-docs verification, not a live API test. We did not measure latency, recall, or citation quality. Prices, credit rules, and limits change; treat the vendors’ live pages as source of truth.
Quick comparison
| Exa | Tavily | |
|---|---|---|
| Official positioning | “The world’s data for AI agents.” Search over Exa’s own index, with token-efficient contents / highlights in the same API. | “Connect your AI agents to the web.” One API for real-time search, plus Extract, Map, Crawl, and Research. |
| Billing model | Pay-as-you-go dollar credits. No subscription and no minimum spend. You load credits and pay per request. | Monthly credit plans, plus optional pay-as-you-go at $0.008/credit after the plan limit. Credits reset on the first day of each month. |
| Free starting point (official) | $10 in credits the day you sign up; the free balance resets to $10 on the 1st of each month (no rollover). Officially “up to 2,500 Lite searches.” First team only. One-time $10 onboarding bonus (does not expire). No card required. | Researcher: 1,000 credits/month, no card. One basic search is 1 credit, so the free month is 1,000 basic (or fast / ultra-fast) searches, or 500 advanced searches. |
| Standard search list price | Lite $4 / 1k requests (~$0.004/call, up to 10 results). Auto (official general-purpose default) $7 / 1k. Extra results above 10 add $1 / 1k results. | Basic, fast, ultra-fast: 1 credit; advanced: 2 credits. At Project rates that is $0.0075 or $0.015 per call; at PAYG, $0.008 or $0.016. Default max_results is 10 (max 20). |
| Known-URL extraction | Contents API: $1 / 1k pages, per content type (text, highlights, or summary each bill separately). | Extract: failed URLs are not charged. Basic = 1 credit per 5 successful URLs; advanced = 2 credits per 5. |
| Research / deeper modes | Deep Search on /search: $12–$15 / 1k requests. Separate Agent API at fixed effort ($0.012–$1.00/run) or metered usage. | Tavily Research: dynamic credits with official floors/caps — mini 4–110 credits/request; pro 15–250. |
| Site map / crawl | Not the center of the product. Use Contents for known URLs; crawling a domain is a different job (see Firecrawl alternatives). | Official Map and Crawl in the same credit system (Crawl = map cost + extract cost). |
| Default rate limits (official) | Team-wide 10 QPS on /search (Deep types 5 QPS; /contents 100 QPS). Buying $1,000 in credits in any 30-day window raises the team limit to 25 QPS for 90 days. | 100 RPM on Development keys; 1,000 RPM on Production keys (Production requires a paid plan or PAYGO). Crawl 100 RPM; Research task creation 20 RPM. |
| Enterprise / compliance (official) | SOC 2 Type II. Enterprise annual plan from the dashboard: $12,000 upfront, includes $12,000 in credits, Zero Data Retention, and 25 QPS. HIPAA / BAA and SSO called out on Enterprise. | Custom Enterprise (custom volume, rate limits, SLAs, security). Official blog (Feb 10, 2026) says Tavily is joining Nebius, and that the API, data policies, and zero data retention remain in place. Confirm current trust/legal pages at publish time. |
How to read this table: List prices are comparable only after you match what one call returns. Official Exa Search docs say text and highlights are included at no extra cost in a search response. Official Tavily Search always returns per-result content snippets; full raw content is an opt-in flag (include_raw_content). This page does not test whether those payloads are equivalent.
Exa — index-first search, billed in dollars per request
Official site: exa.ai · Search product · Pricing · Billing & rate limits
Exa markets a search API over its own index, built for agents. Official product language emphasizes semantic (meaning-based) search, token-efficient highlights, and one API that can be tuned from low-latency search to multi-step research. Documented surfaces include:
- Search —
/searchwith types Lite, Auto (the official default), and Deep variants; official FAQ says a result can include URL, title, publication date, plus highlights, full text, or a summary. Text and highlights are included at no extra cost. - Contents — full page text, highlights, and summaries for known URLs.
- Answer — an LLM answer with citations.
- Monitors — scheduled searches for new events.
- Agent — async deep research, list building, and enrichment (plus optional Connect data providers, billed separately).
- Hosted MCP server, official Python / JavaScript SDKs, and a free-tier mention of MCP access.
Official latency targets on the Search product page (vendor figures, not our measurements): Lite about 450ms at $4 / 1k; Auto about 1 second at $7 / 1k; Deep 4–15 seconds at $12 / 1k. Deep-reasoning is listed as a legacy type at $15 / 1k (12–40 seconds).
Pricing (official)
Exa is pay-as-you-go. There is no monthly subscription and no minimum spend. Usage draws down a prepaid dollar credit balance at the rates below.
| Endpoint / type | Official base price | Notes from the pricing glossary |
|---|---|---|
/search (lite) | $4 / 1k requests | Up to 10 results. Extra results $1 / 1k. Lite has no AI-summary add-on line. |
/search (instant) | $4 / 1k | Legacy type; AI summaries $1 / 1k pages. |
/search (auto, fast) | $7 / 1k | Official general-purpose default (auto). fast is legacy. Extra results and summaries as above. |
Deep Search (deep-lite, deep) | $12 / 1k | Multi-step research / structured outputs. 5 QPS limit. |
Deep Search (deep-reasoning) | $15 / 1k | Legacy type. |
/contents | $1 / 1k pages, per content type | text, highlights, and summary each count. One page with text + highlights = two billable pages. |
/answer | $5 / 1k requests | — |
/monitors | $2 / 1k requests | Extra results and summaries follow the search add-on rates. |
Agent, fixed effort | $0.012 (minimal) to $1.00 (xhigh) per request | Metered auto / ultra bill ACUs and tool calls up to default caps of $5 / $20. |
Free tier and Enterprise (official billing docs):
- Free: $10/month in credits, first team only, resets on the 1st, does not accumulate. One-time $10 onboarding bonus does not expire. Exhausted credits return 402. Officially no payment method required. The free tier does not include Exa Snapshot.
- Default team rate limit: 10 QPS on
/searchand/answer; 5 QPS on Deep types; 100 QPS on/contents; Agent 50 concurrent runs. - Buy $1,000 in credits within any 30-day window → 25 QPS for 90 days (purchase amount, not spend; requalifying resets the window).
- Self-serve Enterprise: $12,000 paid upfront annually, includes Zero Data Retention, $12,000 in credits, and 25 QPS. Custom volume, postpaid invoicing, and higher QPS are sales-led.
- Compliance called out on official pages: SOC 2 Type II, GDPR, CCPA; HIPAA / BAA and ZDR on Enterprise.
Who Exa is for
- Teams that want dollar-denominated, no-subscription search and are fine topping up credits
- Agents that benefit from official highlights / token-efficient excerpts in the search response, without a separate extract call
- Workflows that may grow into Deep Search, Monitors, or the Agent API (list building, enrichment) in the same account
- Buyers who need a published SOC 2 Type II statement and, later, Enterprise ZDR / HIPAA packaging
Who should not pick Exa first
- People who mainly need to map or crawl a whole site — that job is closer to Tavily Map/Crawl or a crawler such as Firecrawl
- Teams that want a fixed monthly credit bundle and a single “1 credit = 1 basic search” mental model
- Workloads that must stay on the free tier with more than about 2,500 Lite searches a month (or fewer Auto/Deep calls — $10 does not stretch as far at $7–$15 / 1k)
- Anyone treating Exa’s published benchmark tables as an independent test — they are vendor-reported; this page does not verify them
Internal link: Exa tool page
Tavily — credit-plan search, with extract, map, and crawl in the same API
Official site: tavily.com · Pricing · Credits & pricing · Search API · Rate limits
Tavily markets itself as a search API for AI agents, with Extract, Map, Crawl, and Research in one credit system. Official homepage copy focuses on live web context for models, production retrieval, and built-in safeguards (PII / prompt-injection / malicious-source language on the marketing site). On 2026-02-10 Tavily published “Tavily is Joining Nebius”; current official branding includes “by Nebius.” That blog says the API, data policies, and zero data retention were intended to stay the same. Confirm the live trust pages if compliance wording matters to a contract.
For this comparison, the documented pieces that matter are:
- Search — depths
basic,fast, andultra-fastcost 1 credit;advancedcosts 2 credits. Defaultmax_resultsis 10 (maximum 20). Results include acontentfield (snippets / chunks). Optional flags includeinclude_answer,include_raw_content(markdown or text), topic filters (general/news/finance), date windows, domain include/exclude, andmax_age_hoursfor cache freshness. - Extract — up to 20 URLs; default output Markdown; failures not charged (see also Firecrawl alternatives for how this sits next to dedicated crawlers).
- Map and Crawl — Crawl is officially mapping cost + extraction cost.
- Research — a separate, variable-credit research task (
miniorpro) with official min/max bounds.
Pricing (official credits docs)
| Plan | Credits / month | Monthly price | Price per credit (docs) |
|---|---|---|---|
| Researcher | 1,000 | Free | — |
| Project | 4,000 | $30 | $0.0075 |
| Bootstrap | 15,000 | $100 | $0.0067 |
| Startup | 38,000 | $220 | $0.0058 |
| Growth | 100,000 | $500 | $0.005 |
| Pay as you go | Usage | $0.008 / credit | $0.008 |
| Enterprise | Custom | Custom | Custom |
Credits reset on the first day of each month, regardless of billing date. Official FAQ: if you exhaust monthly credits, requests stop until reset, or you upgrade / enable paid usage.
Credit costs (official):
| Operation | Official credit cost |
|---|---|
Search basic / fast / ultra-fast | 1 credit per request |
Search advanced | 2 credits per request |
Extract basic | 1 credit per 5 successful URLs |
Extract advanced | 2 credits per 5 successful URLs |
| Map | 1 credit per 10 successful pages (2 per 10 with natural-language instructions) |
| Crawl 10 pages, basic extract (docs example) | 3 credits (1 map + 2 extract) |
| Crawl 10 pages, advanced extract (docs example) | 5 credits |
Research mini | 4–110 credits per request |
Research pro | 15–250 credits per request |
The official Search docs also warn that auto_parameters may set search_depth to advanced, which uses 2 credits. Set search_depth to basic explicitly if you need to avoid that.
The marketing pricing page advertises a startup program with 100k free API credits and free for students. Both are promotional; confirm current eligibility rather than treating them as default plan terms.
The credits documentation prices Search by depth only. It does not list a separate surcharge for flags such as include_raw_content or include_answer. If those flags matter to your bill, confirm on the live credits page before you assume they are free.
Rate limits and freshness caveats (official)
- Development keys: 100 RPM. Production keys: 1,000 RPM. Official docs: Production access requires an active paid plan or PAYGO.
- Crawl: 100 RPM on both key types. Research creation: 20 RPM (status polling uses the default limit).
- On non-paying plans, search requests that set
max_age_hoursare served on a flex tier. Under load they may return 434 Flex tier capacity exceeded. Official workaround: retry, or omitmax_age_hoursand use cache.
Who Tavily is for
- Agent and RAG builders who want search + extract + optional site map/crawl in one credit balance
- Teams that prefer a monthly plan (or a hard 1,000-credit free cap) over managing a prepaid dollar wallet
- Jobs that are mostly known-URL Extract, where basic extract is 5 URLs per credit
- Users who need official news / finance topic switches,
include_answer, or raw Markdown on search results
Who should not pick Tavily first
- Buyers who want no monthly plan and a published per-1k dollar menu like Exa’s Lite/Auto table
- Workloads that must live-fetch on the free plan during peak load — official flex-tier 434 behavior applies when
max_age_hoursis set - People choosing on search quality scores from either vendor’s landing page — this page does not validate Tavily’s published evals either
- Teams whose real problem is open-source / self-hosted crawling — Tavily is a hosted API
Internal link: Tavily tool page
How to choose (decision guide)
-
Is the job “search the web for an agent,” or “read / crawl URLs you already have”?
Both APIs search. If the center of the design is known URLs or a whole docs site, Tavily Extract / Map / Crawl is the more complete official surface in this pair. Exa Contents covers known URLs at $1 / 1k pages per content type. For a dedicated crawler comparison, use Firecrawl alternatives. -
Do you want dollars-per-request or a monthly credit bundle?
Exa: load credits, pay $4–$7 / 1k standard searches (more for Deep). Tavily: pick a plan (or PAYG at $0.008/credit) and spend 1 or 2 credits per search. A rough list-price check for 1,000 basic searches: Exa Lite $4; Exa Auto $7; Tavily Project $7.50; Tavily Growth $5; Tavily PAYG $8. That is billing math only — it is not a quality ranking, and it assumes you are comparing a 1-credit Tavily search to an Exa search with the default result cap. -
What does the free tier actually buy?
Exa: $10/month (first team), officially up to 2,500 Lite searches, no rollover, plus a one-time $10 bonus. Tavily: 1,000 credits/month, i.e. 1,000 basic searches or 500 advanced, with a hard stop when the balance hits zero unless you enable paid usage. -
How hard is the rate limit in practice?
These units are not identical (QPS vs RPM). Exa’s default 10 QPS is a higher sustained cap than Tavily Development (100 RPM); Tavily Production (1,000 RPM) requires paid/PAYGO and is still a different shape than Exa’s 25 QPS unlock after a $1,000 credit purchase. Match the official limit to your key type and burst pattern. -
Do you need research, monitors, or enrichment?
Exa documents Deep Search, Monitors, and a priced Agent API. Tavily documents Research with a wide per-request credit band (4–250 credits). Price a realistic research volume on the live pages before you assume either “research” SKU is a casual add-on. -
Do vendor benchmarks decide this?
No. Exa and Tavily both publish first-party evals (and Exa cites third-party indexes on its product page). This article does not rerun them, and the two vendors do not use a single shared public table against each other here. Choose on documented product fit and billing, then test on your queries.
Who should not use which
| Situation | Prefer | Avoid as primary |
|---|---|---|
| You want prepaid USD, no monthly plan, and a per-1k search menu | Exa | Treating Tavily PAYG as the same pricing shape (it is still credit-denominated at $0.008) |
| You want search + site map/crawl in one vendor | Tavily | Exa Search alone |
| You only extract Markdown from a list of URLs | Tavily Extract (or Exa Contents if you already use Exa) | Standing up the other vendor’s full search stack just for that |
| You need the official SOC 2 Type II + Enterprise ZDR/HIPAA packaging described on the marketing/docs site | Exa’s documented Enterprise path | Assuming Tavily’s Nebius announcement automatically equals the same certificates — check Tavily’s live trust pages |
| Free-tier volume above ~1,000 basic searches / month, and Lite quality is enough | Exa Free ($10, up to 2,500 Lite) | Tavily Researcher alone |
| You must pin live (uncached) results on a free plan | Exa, or a paid Tavily plan | Tavily free + max_age_hours during flex-tier load (official 434) |
| Your job is really “crawl this domain into Markdown for RAG” | Firecrawl or Tavily Crawl | Using either search API as a crawler substitute without reading crawl/extract docs |
When neither fits
Adjacent official products that sometimes show up in the same shortlists: Firecrawl (site-to-Markdown crawl; see Firecrawl alternatives), Parallel, traditional search/CSE APIs, or a self-hosted crawler. Use those when the job is raw browsers, enterprise proxy networks, or “I already have the URLs and only need a crawler.”
FAQ
Is Exa cheaper than Tavily?
It depends on search type and plan, not on a single sticker. Official list math for 1,000 basic-depth searches can make Exa Lite ($4) cheaper than Tavily Project ($7.50) or PAYG ($8), while Exa Auto ($7) is close to Tavily Project, and Tavily Growth ($5 / 1k credits) sits between Lite and Auto. Advanced Tavily searches cost two credits. Exa Deep Search and Tavily Research are priced on different axes and should not be compared with that 1,000-search shorthand. Re-run the math on the live pricing pages for your mix of search, extract, and research calls.
Does Tavily replace Exa Search, or the other way around?
Not as a 1:1. Both are agent search APIs. Exa is index- and search-first, with Contents, Answer, Monitors, and Agent around that core, billed in dollars. Tavily is search plus extract/map/crawl/research in one credit wallet, sold as monthly plans. Pick the official surface that matches the workload.
Can I use both?
Yes. A common official-docs-compatible split is Exa or Tavily for query search, and Tavily Extract / Firecrawl / Exa Contents for pages you already have. Two vendors means two bills and two failure modes; only do it if the endpoints differ enough to matter.
What did “Tavily is joining Nebius” change?
Tavily’s official blog (2026-02-10) said it was joining Nebius and that day-to-day API use, data policies, and zero data retention were not supposed to change. Official pages now show “by Nebius.” This article does not track contract, entity, or closing details beyond those official pages. If you are renewing an enterprise agreement, read the current legal/trust docs.
Do ChatGPT / Claude browsing tools replace these APIs?
No. Those are product features inside a chat app. Exa and Tavily are APIs you call from your own agents, with documented pricing, rate limits, and extract/crawl options.
Sources & last checked
Checked: 2026-11-10, Asia/Shanghai (UTC+8). Official-docs verification only; no live search-quality or latency test.
No traffic estimates, Wikipedia, or Product Hunt links are used as Website sources on this page. Vendor benchmark scores are not reproduced as our results.