Exa
Exa Labs · Ranked #3 of 12 in AI Candidate Sourcing Tools
Neural web search API with a people search category (1B+ public profiles, launched Dec 2025) and Websets for building entity lists; self-serve with a monthly $10 free credit, Python and JavaScript SDKs and a hosted MCP, but no structured title/seniority filters or verified contact data by default.
Developers and AI agents that want semantic people discovery across the open web, not exact-field recruiter filters

Overview
Exa is a search API built on its own embedding index. For sourcing, its relevant surfaces are the people category on /search (announced December 19, 2025, covering 1B+ public profiles and replacing the older linkedin category), Websets (asynchronous jobs that build lists of people, companies or papers against natural-language criteria) and the Agent API, which bills email and phone enrichment separately ($0.02 and $0.07). It is a general web retrieval system rather than a candidate database: a competing vendor argues it lacks structured title, seniority and geography filters and returns duplicates, which is a vendor claim we could not test.
On the developer surface the evidence is strong. The docs include a quickstart with Python, JavaScript and cURL, an OpenAPI spec, a changelog and an llms-style index; Python (exa-py) and JavaScript (exa-js) SDKs; a hosted MCP at mcp.exa.ai that works without an API key; and an agent skill. Pricing is published and pay-as-you-go ($4-$7 per 1,000 searches, $12-$15 for deep search, $1 per 1,000 content pages) with $10 of free credit that resets monthly and needs no card. Websets and people search have no dedicated price on the pricing page, and standard rate limits are not stated.
Exa runs a public status page (status.exa.ai) showing 99.97% for the Search API, 99.99% for Websets, 99.92% for the Agents API and 100% for MCP; third-party monitors recorded 22 incidents in 90 days (about 96.3% availability), including a roughly two-hour elevated-error event on filtered searches on May 23. SLAs exist only on enterprise contracts. No G2 or Trustpilot review corpus was found.
How this score is derived
The APIbenchmarks Index is a weighted sum of four dimensions, each scored on an absolute 0–100 reference scale. See the methodology for every mapping.
| Dimension | Score | Weight | Contribution |
|---|---|---|---|
| Documentation & DXexa.ai/docs offers a quickstart in Python, JavaScript and cURL, an OpenAPI spec, a changelog and separate Websets and MCP sections; strong. The people search page and Websets reference were not fully read, and standard rate limits are not documented, so scored below the top band. | 86 | 30% | 25.8 |
| ReliabilityPublic status page with 99.97% (Search), 99.99% (Websets), 99.92% (Agents) and 100% (MCP) uptime, but third-party monitors count 22 incidents in 90 days (about 96.3%) including a ~2h filtered-search degradation on May 23; SLAs are enterprise-only. Scored conservatively because the two sources disagree. | 80 | 25% | 20.0 |
| Ecosystem & SDKsOfficial Python (exa-py) and JavaScript (exa-js) SDKs, a hosted keyless MCP server, an agent skill and a Websets MCP; no Go/Java SDKs and no native ATS integrations. | 84 | 25% | 21.0 |
| AccessibilityHighest in the category set: $10 monthly free credit with no card, published pay-as-you-go pricing and a self-serve dashboard key. Websets and people search have no separately listed price and enterprise limits need sales. | 90 | 20% | 18.0 |
| APIbenchmarks Index (ABI) | 84.8 | ||
Table 1. Derivation of the ABI for Exa. Contribution = score × weight; the index is their sum.
Exa vs the AI Candidate Sourcing Tools field
APIbenchmarks Index · higher is better
Build quality vs accessibility
Exa (dark dot) among AI Candidate Sourcing Tools APIs
At a glance
- Vendor
- Exa Labs
- Pricing model
- Pay-as-you-go per 1,000 requests; enterprise custom
- Free tier
- $10 credits for new accounts, reset monthly, no card required
- Official SDKs
- 5 languages
Pricing
| Free credits | $10 / month | New accounts; resets on the 1st; no card required. |
| Search | $4-$7 per 1k requests | Lite/instant $4, auto/fast $7; $1 per 1k results above 10. |
| Deep search | $12-$15 per 1k requests | deep-lite and deep $12, deep-reasoning $15. |
| Enterprise | Custom | Volume discounts, custom QPS, SLAs, Zero Data Retention. |
Key features
- •Semantic search with people category
- •Websets (entity lists with criteria verification)
- •Deep search tiers
- •Contents extraction
- •Agent API with email/phone enrichment
- •Hosted MCP server
- •Monitors
Official SDKs
Strengths & trade-offs
- +$10 free credit that resets monthly, no card needed
- +Published pay-as-you-go pricing
- +Python and JavaScript SDKs plus a hosted MCP without an API key
- +People search over 1B+ public profiles and Websets for entity lists
- +Public status page with per-component uptime
- +Agent API can enrich email and phone
- –No structured title/seniority/geography filters (vendor-reported)
- –Websets and people search have no dedicated public price
- –Third-party monitors show frequent short incidents
- –No standard rate limits published
- –Not a candidate database; results are web pages and profiles
What developers say
No G2 or Trustpilot corpus found
No independent review corpus was located; the only critical view comes from a competing vendor and third-party monitors show frequent short incidents against a strong public status page.
“A competing data vendor argues Exa returns semantically similar pages without structured title or seniority filters (paraphrased, vendor claim).”
Key figures
| Free credits | $10, resets monthly | Exa pricing ↗ |
| Search API uptime | 99.97% | Exa status page ↗ |
| Websets uptime | 99.99% | Exa status page ↗ |
| Incidents, last 90 days | 22 (about 96.33% uptime) | Pingoru (third party) ↗ |
| People index | 1B+ public profiles | Exa changelog ↗ |
Compare Exa head to head
Sources
- https://exa.ai/docs/introduction
- https://exa.ai/pricing
- https://exa.ai/docs/changelog.md
- https://docs.exa.ai/websets/faq
- https://status.exa.ai
- https://pingoru.io/providers/exa-ai
- https://crustdata.com/blog/why-exa-doesnt-work-for-structured-people-search
Figures last verified 2026-10-22. Spotted an error? corrections@apibenchmarks.com
