A calibrated score on every result
Not an opaque rank — a probability. 0.94 means relevant. 0.51 means we’re not sure, and we say so, so your agent doesn’t have to guess.
Our own continuously refreshed index, a calibrated relevance score on every result, and drop-in compatibility with the Exa and Tako SDKs. Keep your code. Cut your search bill by 80%.
$1.40 per 1,000 searches · Exa and Tako list at $7
Sample outputRelevance is a calibrated probability. Between 0.35 and 0.65 we say “undecided” instead of guessing.
Speaks the SDKs you already ship
The numbers
Public list prices, and measurements from our own test suite. No cherry-picked benchmarks.
| typesearch | Exa | Tako | |
|---|---|---|---|
| Search | $1.40 | $7 | $7 |
| Deep search | $2.40 | $12–15 | $12 |
| Contents · per 1k pages | $0.20 | $1 | $1 |
Prices: exa.ai/pricing and tako.com/pricing, September 2026. Speed and compatibility: our internal test suite. Methodology on request.
Search API
Not an opaque rank — a probability. 0.94 means relevant. 0.51 means we’re not sure, and we say so, so your agent doesn’t have to guess.
Savings accounts: what changes for depositors
Rate decision: what it means for mortgages
Headlines propose; the article decides. In Normal and Deep modes we open the top candidates and verify them. What doesn’t hold up never reaches your context window.
// Mentions a specific figure? "figure": { probability: 0.93 }, // Main actor "actor": { choice: "central bank", confidence: 0.88 }, // Impact on consumers (1–5) "impact": { score: 3.6, confidence: 0.81 }
Ask typed questions — yes/no, multiple choice, 1–5 scores — and get calibrated answers on every result, in the same call. No second LLM pass. No parsing prose.
Do the math
Slide to your monthly volume. List prices, no asterisks.
Published list prices on exa.ai/pricing and tako.com/pricing, checked September 2026, before free tiers. Deep search uses Exa’s lowest deep tier.
You keep
$67,200a year
The index
We crawl, refresh and rank our own index. Sources are re-read every few minutes from their news sitemaps and feeds, so what’s published at 9:02 is searchable by 9:05 — in the language it was written.
Beyond ten blue links
Positive, neutral or negative about what you asked — per result and per source. One scale for news, blogs and social.
The same wire story told by ten outlets collapses into one result, with the duplicates attached.
Coverage by day and by source, and who broke it. Computed from the index at no extra cost.
Up to three verbatim excerpts from different sources. We quote; we don’t write. Zero hallucinated summaries.
Set a budget per request. We return what fits and mark the response incomplete — never a surprise invoice.
Search any site right now: its homepage, the sections worth reading and its own search box, in parallel.
Migration
One compatibility layer per provider: same parameters, same response shape, same errors. Point your official SDK at us and ship.
from exa_py import Exa-exa = Exa(api_key=EXA_API_KEY)+exa = Exa(api_key=TYPESEARCH_API_KEY, base_url="https://api.typesearch.ai/compat/exa")results = exa.search("lithium royalties in Chile", num_results=10, type="fast")similar = exa.find_similar(results.results[0].url, exclude_source_domain=True)pages = exa.get_contents([r.url for r in results.results], highlights=True)
Production
/v1 only changes by adding. Anything breaking ships as /v2, with notice.
Daily quotas and per-minute limits on every key, visible in real time through the usage API.
RFC 9457 problem details, stable error codes and a request ID on every response.
We store a hash, never your key. Rotate and revoke anytime.
The contract is generated from the same schemas that validate every request.
JSON or server-sent events on every search endpoint, with partial results as they land.
Volume pricing, custom source coverage, dedicated support and SLAs. Tell us what you pay today — we’ll show you what you’d pay here.
Yes, on list price. Search is $1.40 per 1,000 requests; Exa and Tako both list $7. Deep search is $2.40 against $12–15, and contents $0.20 against $1 per 1,000 pages. Put your own volume in the calculator above.
It shouldn’t. We run the official SDKs — Python and JavaScript — against our compatibility layers: 25 of 25 test cases pass by changing only the base URL and the key. Anything we don’t support is accepted and reported in a response header, not thrown as an error.
News and editorial content, crawled and refreshed by us every few minutes, with deep coverage of local and regional press in Latin America and Europe. The sources endpoint returns the live list.
No, on purpose. We return evidence: ranked, verified, with verbatim excerpts. If you need prose, your model writes it — grounded in results that can’t be hallucinated.
Two honest limits: we return short excerpts instead of full page text, and we don’t offer people, company or academic-paper indexes. If those are core to your product, keep them where they are and move the rest of your search to us.
Yes. Every new account gets $5 in free credit, about 3,571 searches on any endpoint, valid for 30 days. No card needed. Your first top-up gets 20% extra, up to $20.
Get a key in a minute. Point your SDK at us. Watch the invoice shrink.