Exa is live · free plan; usage-based paid tier · last verified 13 Nov 2026 · viability healthy
Exa
Search API for AI agents that returns token-efficient highlights or schema-shaped JSON instead of blue links.
Exa 3.0 is the retrieval API to beat on hard agent queries. The vendor's own charts put it at 85.5% on WebCode against Perplexity (73.9%), Brave (69.4%) and Parallel (76.3%), and it reports leadership on BrowseComp, DSQA and WideSearch. Pick it when your agent needs token-efficient excerpts (Highlights), grounded citations (Deep at $12/1k requests) or structured company data across 70M+ companies. Look at Brave or Perplexity if you want a cheaper general index; keep a plain vector store if your corpus is internal. Benchmark on your own queries before committing — published evals won't tell you whether it wins on your data.
Verdict generatedVerified 19h ago · liveness 87/100 · cite: rightaichoice.com/tools/exa
- Engineering teams wiring retrieval into coding agents
- Research assistants where every answer must carry checkable citations
- Sales and GTM teams enriching company and people lists at scale
- Voice and chat agents that need sub-second search during a turn
- Anyone who wants a graphical search engine to browse results themselves
- Pipelines that need raw rendered HTML rather than token-efficient excerpts
- Teams that need a flat monthly bill instead of per-request pay-as-you-go
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Exa if your pipeline needs raw rendered HTML or an interface a human browses, or if you need a flat predictable monthly bill rather than per-request charges across Search, Contents, Monitors and Agent.
Each result beyond the first 10 adds $1 per 1k requests on lite, auto and deep, so wide result counts quietly raise the per-search bill.
Starter's $10 monthly credit plus a $10 onboarding bonus covers prototyping at 10 Search QPS; Developer is pure pay-as-you-go with no commitment and up to 25 Search QPS; Enterprise is custom. Per-request rates from $4/1k are competitive with general search APIs, but teams wanting flat monthly bills should compare against a subscription search tool, and Zero Data Retention starts at a $12,000 annual commitment in API credits.
In short
Exa — Search API for AI agents that returns token-efficient highlights or schema-shaped JSON instead of blue links. Best for Engineering teams wiring retrieval into coding agents, Research assistants where every answer must carry checkable citations, Sales and GTM teams enriching company and people lists at scale. Free to start; it also has paid plans whose price we could not confirm — see the pricing table.
What is Exa?
Exa — Search API for AI agents that returns token-efficient highlights or schema-shaped JSON instead of blue links. Best for Engineering teams wiring retrieval into coding agents, Research assistants where every answer must carry checkable citations, Sales and GTM teams enriching company and people lists at scale. Free to start; it also has paid plans whose price we could not confirm — see the pricing table.
Exa is a search API your agent calls instead of a search box you browse. One request against Exa's own index — the vendor says it holds 100 billion documents in its vector DB and tracks 2 trillion URLs — comes back as structured JSON your model reads directly: page text, token-efficient highlights, or schema-shaped enrichments across 70M+ companies. Exa 3.0 shipped October 19, 2026 as the vendor's largest-ever jump in search quality; its published charts put accuracy at 85.5% on WebCode against Perplexity (73.9%), Brave (69.4%) and Parallel (76.3%), and the vendor says it leads BrowseComp, DSQA and WideSearch. The recommended search types are now lite, auto and deep. Lite runs at $4 per 1k requests for up to 10 results, serves page content from Exa's cache and never fetches live, and returns a 400 naming any parameter it does not support. Auto answers in about 1s; Deep takes roughly 10s with grounded citations. Search responses carry warnings and notices arrays so your agent can see when a field had no effect. Billing is per request: Search from $4/1k, Contents at $1/1k pages per content type, Monitors at $2/1k, Agent runs at $0.012–$1.00 by effort mode, with Exa Connect pulling paid third-party data providers on top. It suits engineering teams wiring retrieval into coding agents, research assistants, voice agents and GTM enrichment.
Key features
- Exa 3.0 search with lite, auto and deep search types
- Lite search at $4/1k requests, 300ms, up to 10 results, cached content only
- Auto search at $7/1k requests for parameters lite does not support
- Deep Search at $12/1k requests, ~10s, grounded citations
- Legacy fast, instant and deep-reasoning search types kept at current prices
- Warnings and notices arrays on search responses explain ignored fields
- Highlights extraction model returns only relevant excerpts from each page
- Dynamic Highlights research preview (49% better token efficiency in single-turn RAG)
- Contents endpoint returns full page text and highlights at $1/1k pages per content type
- Configurable livecrawl policies on Contents
- Agent API with effort modes from Minimal $0.012 to X-high $1.00 per request
- Agent Ultra usage-based mode with $20 default per-run cap and 5-minute to 3-hour budgets
- Monitors API runs searches on a cadence and delivers fresh events via webhooks
- Structured extraction via output schema across 70M+ companies
- Exa Places: structured search over 211M locations
Behind the verdict
Exa's pitch is narrow and honest: it is not a search engine you look at, it is a search endpoint your agent calls. That framing shapes everything worth knowing before you sign up. Strengths start with output shape. The Contents endpoint returns full page text and highlights at $1 per 1k pages per content type, and the Highlights extraction model is trained to return only the relevant excerpt — the vendor claims 90% token reduction on its Boeing example and says up to 25T tokens a week passed to models are highlights-derived excerpts. If your context budget is the binding constraint, that is a real saving, not a rounding error. Second, structured output: pass an output_schema and get typed JSON back (the docs example pulls company, CEO, founded_year and headquarters for Anthropic), and Exa markets enrichment across 70M+ companies plus Exa Places over 211M locations. Third, latency tiers you can actually route on: Lite at 300ms, Auto at 1s, Deep at ~10s, with Agent runs priced $0.012 (Minimal) to $1.00 (X-high) per request and Ultra billed at standard usage rates under a default $20 per-run cap. Weaknesses are mostly about cost control and scope. Billing is per request per endpoint — Search from $4/1k, Contents $1/1k pages per content type, Monitors $2/1k, Agent runs at usage rates — plus $1/1k for each result above 10. Nothing aggregates into a flat bill, so a chatty agent multiplies your invoice silently. Lite serves cached content only and never fetches live, so anything time-sensitive has to go to Auto or Deep. Zero data retention is not self-serve: the pricing page lists it as a $12,000 annual commitment in API credits with 25 Search QPS, upgradeable from the dashboard. Category churn is another watch item — the pdf, github and tweet search categories are being deprecated, startCrawlDate/endCrawlDate are now ignored for all teams, and the publication category replaced research paper. Where it fits: coding agents needing current docs and repos, research assistants where citations must be checkable, voice agents that need sub-second grounding mid-turn, and GTM list-building where you want typed company and people fields rather than scraped HTML. Where it doesn't: if you want to eyeball results yourself, or your corpus is internal and a plain vector store already answers it, Exa adds a vendor and a per-request bill for little gain. Also skip it if your pipeline needs raw rendered HTML — you get token-efficient excerpts, not a browser.
Models under the hood
as of 2026-11-07
Limitations
- Pay-as-you-go billing stacks across endpoints: Search from $4/1k requests (lite, up to 10 results, plus $1/1k results above 10), Contents at $1/1k pages per content type, Monitors from $2/1k requests, and Agent runs at $0.012–$1.00 with Ultra defaulting to a $20 per-run cap and 5-minute to 3-hour budgets.
- The free Starter tier is capped at 10 Search QPS and 50 agent concurrency; up to 25 QPS needs Developer, and custom indexes, up to 1,000 results per search, SSO/SCIM and a dedicated Slack channel require Enterprise.
- Lite only serves cached page content — it never fetches a page live — and an unsupported parameter returns a 400 naming it.
- Deep and Ultra research runs are asynchronous.
- Dynamic Highlights still requires the Exa-Beta header, startCrawlDate/endCrawlDate are now ignored for all teams, and the pdf, github and tweet search categories are being deprecated.
as of 2026-11-13
Researching Exa? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Exa still active?
Exa is live · free plan; usage-based paid tier · last verified 13 Nov 2026 · viability healthy.
Viability score
How well maintained and how widely used is Exa? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: November 2026
How we score →What's new in Exa
Checked todayAcross the latest 4 updates: 3 feature updates and 1 launch.
Lite search, recommended search types, and search warnings
Lite becomes a recommended search type at $4 per 1k requests with cached content only, and /search responses now carry warnings and notices arrays that explain ignored fields to both developer and model.
Introducing Exa 3.0
Exa 3.0 ships as the vendor's largest ever improvement to search quality, alongside the lite search type and upgraded highlights.
Agent Ultra effort mode for Exa Agent runs
Set effort: ultra for large list building and hard-to-verify criteria, metered at standard Agent rates with a default $20 per-run cap and a 5-minute to 3-hour duration budget.
Dynamic Highlights research preview
Dynamic Highlights selects excerpts across the complete result set, reporting about 49% better token efficiency in single-turn RAG; requests must send the Exa-Beta: dynamic-highlights-2026-08-28 header.
What do users say about Exa?
What people actually say about Exa — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 7 sources (Hacker News, YouTube, Product Hunt, App Store, Bluesky, Stack Overflow, Lemmy) · researched Nov 13, 2026.
Weighted by the 78 posts each of 7 sources contributed.
- +Search from $4 per 1k requests undercuts Kagi's $12 per 1k considerably
- +Named alongside Parallel as a top-quality option on a public NEEDLE benchmark
- +Returns structured JSON with highlights, cutting token spend versus raw page scraping
- +Product Hunt users describe it as a daily workflow tool they'd recommend
- +Generous roughly $10 monthly free allowance noted by casual users
- −Almost no public commentary on uptime, support or failure modes at scale
- −Lite tier's cached-only results make it unusable for live or breaking information
- −Advanced skill level and three search types add setup decisions for new users
- −Independent benchmark validation exists only as a single community comment
- −Agent run pricing ($0.012–$1.00) is harder to budget than flat search requests
- • Livecrawl and third-party data via Exa Connect add charges on top of base search
- • Deep search at $12/1k plus agent runs can escalate quickly in autonomous loops
How we verified this
Every listed fact about Exa — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 13 Nov 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified Exa 22 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed setup time, pricing fit, switching out, switching in, hidden costs, description, who should skip it, features, summary
- — changed pricing page
- — changed summary, pricing fit, description, our verdict, features, who should skip it, hidden costs, switching in, setup time, switching out, models
- — changed switching out, switching in, summary, description, who should skip it, limitations, features, our verdict, setup time, hidden costs, pricing fit
- — changed pricing page
- — changed use cases, switching in, description, our verdict, summary, who should skip it, features, who it suits, hidden costs, integrations, switching out, setup time, limitations, pricing fit
Showing the 6 most recent of 22 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Exa cost in 2026?
Exa: Free to start; it also has paid plans whose price we could not confirm — see the pricing table. Pricing last verified 13 Nov 2026.
Plans compared
Every published Exa tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Starter | Free | — | Solo developer or small team prototyping an agent who wants to test every endpoint before committing a payment method.Adds: Free entry point: $10 credits monthly plus a $10 onboarding bonus, all endpoints, MCP server and Claude Connector, 10 Search QPS and 50 agent concurrency. |
| Developer | Usage-basedVendor: “Pay as you go” | usage | Production teams shipping an agent that needs more throughput and self-serve access to third-party data providers.Adds: Adds pay-as-you-go billing with no commitment, standard email support, up to 25 Search QPS, SOC 2 Type II and self-serve Exa Connect providers. |
| Enterprise | Custom | quote | Organizations with compliance requirements or search volume that outgrows the standard QPS limits.Adds: Adds zero data retention, HIPAA with BAA, SSO and SCIM, custom indexes, up to 1,000 results per search, custom QPS and monthly invoicing. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Exa's pricing actually pencils out — and where peers do it cheaper.
Starter's $10 monthly credit plus a $10 onboarding bonus covers prototyping at 10 Search QPS; Developer is pure pay-as-you-go with no commitment and up to 25 Search QPS; Enterprise is custom. Per-request rates from $4/1k are competitive with general search APIs, but teams wanting flat monthly bills should compare against a subscription search tool, and Zero Data Retention starts at a $12,000 annual commitment in API credits.
Has Exa's pricing changed?
No plan price for Exa has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Exa?
Pick Exa if you are Engineering teams wiring retrieval into coding agents, Research assistants where every answer must carry checkable citations, Sales and GTM teams enriching company and people lists at scale, Voice and chat agents that need sub-second search during a turn.
Look elsewhere if you are Anyone who wants a graphical search engine to browse results themselves, Pipelines that need raw rendered HTML rather than token-efficient excerpts, Teams that need a flat monthly bill instead of per-request pay-as-you-go, Workflows that point Agent Ultra at long lists without setting a cost ceiling in code.
Skip it if Skip Exa if your pipeline needs raw rendered HTML or an interface a human browses, or if you need a flat predictable monthly bill rather than per-request charges across Search, Contents, Monitors and Agent.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Exa actually fits — and what changes day-one when you adopt it.
Wire type: "lite" into the agent loop so doc and repo lookups return cached highlights at 300ms, and fall back to auto only when a filter lite does not support is needed.
Outcome: Lower latency per turn and a smaller context bill, with the warnings array surfacing any ignored field before it reaches production.
Call type: "deep" with an output_schema for the fields you need, then render the grounded citations next to each answer.
Outcome: Answers arrive as typed JSON plus checkable sources, so reviewers can verify each claim without re-running the search.
Use Exa Agent with effort: "ultra" and budget.maxCostDollars set to a hard ceiling to build a qualified company list, then push enriched records into the CRM.
Outcome: A list built under a known spend cap, with per-run accounting visible before anything hits the CRM.
Use cases
- Ground a coding agent with current docs, package references and changelogs inside a latency-sensitive loop.
- Turn a query like 'AI startups in healthcare' into a structured company list with CEO names and founding years.
- Add neural retrieval to a RAG pipeline for better grounding than keyword search.
- Run Monitors that watch for new web content matching a query and fire webhook alerts on a cadence.
- Enrich CRM records with company and people data via output schema.
- Build a deep research assistant that synthesizes multiple sources with grounded citations.
- Give a voice agent current information at sub-second latency using lite search.
- Track academic publications across 350M papers with enriched author and funding profiles.
Setup time & first value
How long it actually takes to get something useful out of Exa — broken out by persona, not the marketing-page minute.
With the free Starter credits and no payment method required, an engineer can hit the /search endpoint from the docs quickstart in minutes. Expect under an hour for a working lite or auto prototype, and a day or so to tune schemas, livecrawl policies and per-run budgets before production traffic.
Switching to or from Exa
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From Brave Search API: map your query filters to lite's focused parameter set, and move keyword ranking calls to type: "auto" where lite returns a 400.
- →From Perplexity API: swap the chat-style call for type: "deep" with output_schema so structured fields and grounded citations come back in one request.
- →From a self-hosted vector store: keep it for internal corpora and route only public-web retrieval to Exa Search with highlights enabled.
- →From Tavily or SerpAPI: replace blue-link parsing with contents.highlights and read the notices array instead of post-processing result HTML.
- ↗To Brave Search API: use it for a cheaper general index, then re-add Exa later for highlights and structured extraction.
- ↗To a plain vector store: point retrieval at your own corpus when results only ever come from internal documents.
- ↗To Perplexity: move here if you want a chat-style answer surface rather than an API returning JSON to your agent.
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Exa”, and we withheld 5: 5 could not be judged, because “Exa” is a single word that other videos use for other things. Showing the 1 we can prove is about Exa.
Official links
What are the best Exa alternatives?
The closest alternatives to Exa we list are Tavily, Spider Cloud and Firecrawl. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Tavily | Freemium | Pricing not published | — |
| Spider Cloud | Paid | Pricing not published | — |
| Firecrawl | Freemium | Pricing not published | — |
Alternatives to Exa
View all
Spider Cloud
Web scraping, crawling and search API that returns clean markdown or JSON for RAG pipelines and AI agents.
Tools that pair well with Exa
Common stack mates teams adopt alongside Exa, with the specific reason each pairing earns its keep.
Tavily
A web search API that grounds AI agents with live, structured web context
Spider Cloud
Web scraping, crawling and search API that returns clean markdown or JSON for RAG pipelines and AI agents.
Firecrawl
Firecrawl is a web data API that scrapes, searches, and interacts with live pages to feed LLM-ready Markdown and JSON into AI agents.
Exa vs its closest rival
Exa's closest head-to-head is with Firecrawl.
Our Exa vs Firecrawl verdict written May 12, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
Frequently asked questions
Categories
For makers of Exa
- Get a Verified Listing for Exa ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks of the review starting.
Used Exa? Help shape our editorial sentiment research.