NewYour Exa or Tako SDK, up to 7× cheaper.

Compared with Exa (exa.ai, formerly Metaphor)

vsExa

The Exa alternative, 2.1⁠–⁠7× cheaper

News search for 2.1–7× less than Exa, depending on the mode — $1.00 for ultra, which judges headlines only, and $1.40 for fast against $7, $5.60 against $12–15 in deep — with the exa-py and exa-js SDKs you already have.

Get $5 freeThe whole migration

The same work, mode by mode

  • Ultra search · headlines only

    $1.00

    Exa: $7−85%

    Search, text and highlights of up to 10 results included

  • Fast search

    $1.40

    Exa: $7−80%

    Search, text and highlights of up to 10 results included

  • Normal search · reads the top 4

    $2.20

    Exa: $7−68%

    Search, text and highlights of up to 10 results included

  • Deep search · reads 8

    $5.60

    Exa: $12–15−53–62%

    Deep Search (or Deep-Reasoning)

  • Similar · ultra, fast and normal

    $2.20

    Exa: $7−68%

    Search (findSimilar is deprecated, with no published price)

  • Similar · deep, reads 8

    $4.60

    Exa: $7−34%

    Search, text and highlights of up to 10 results included

  • Contents · per 1k pages

    $0.20

    Exa: $1−80%

    Contents, text, per 1k pages

  • Contents with a query · per 1k pages

    $0.40

    Exa: $2−80%

    Contents, text + highlights, per 1k pages

List prices per 1,000 requests (contents, per 1,000 pages), against the cheapest product Exa sells for the same work.

Checked September 25, 2026 · Sources: exa.ai/pricing, exa.ai/docs/changelog

What stays the same

Your SDK, your parameters, your response parsing. Here is how each Exa parameter maps.

Exa parameterStatusOn typesearch
query, numResultsSameSame. Up to 50 results.
typeMappedinstant → Ultra · fast, auto, neural, keyword, hybrid → Fast · deep → Deep
includeDomains, excludeDomainsSameSame, up to 20. *.domain.com works too.
startPublishedDate, endPublishedDateSameSame.
includeText, excludeTextMappedApplied to title, standfirst and highlights.
contents.highlightsSameVerbatim excerpts the model picks for your query, billed per page read ($0.40 per 1,000). Exa includes contents for the first 10 results in its search: here a fast search with highlights of 10 comes to $5.40 per 1,000, against $7.
contents.textLimitedA short excerpt, never the full page, billed per page ($0.20 per 1,000).
contents.summaryNot supportedWe don’t write summaries. Use essentials instead.
/findSimilar, /contentsSameSame endpoints, same shapes. Exa has deprecated findSimilar; ours keeps working.

What you gain

  • A calibrated relevance probability on every result
  • Top results read and verified before they’re ranked
  • Typed questions answered per result, in the same call
  • Tone, same-story dedupe and who-published-first
  • Up to five queries per request
  • Hard cost caps per request

When Exa is still the better fit

We’d rather you stay than churn in a month. Keep Exa for these:

  • You need full page text for every result.
  • You rely on people, company or research-paper indexes.
  • You want the API itself to write answers or summaries.

Migration

The whole 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)
beforeafter

Switching from Exa

What is the best Exa alternative for news search?

For agents that search news and the open web, typesearch does the same job 2.1⁠–⁠7× cheaper, depending on the mode: from $1.00 per 1,000 searches, against $7 at Exa. Every result carries a calibrated relevance probability, and in normal and deep the top ones are read and verified before they’re ranked.

Can I switch from Exa without rewriting my code?

Yes. Keep exa-py or exa-js and point it to https://api.typesearch.ai/compat/exa with a typesearch key: the endpoints, parameters and response shapes stay the same. The migration is two lines.

How much cheaper is typesearch than Exa?

Ultra search · headlines only: $1.00 against $7 (Search, text and highlights of up to 10 results included). Fast search: $1.40 against $7 (Search, text and highlights of up to 10 results included). Normal search · reads the top 4: $2.20 against $7 (Search, text and highlights of up to 10 results included). Deep search · reads 8: $5.60 against $12–15 (Deep Search (or Deep-Reasoning)). Similar · ultra, fast and normal: $2.20 against $7 (Search (findSimilar is deprecated, with no published price)). Similar · deep, reads 8: $4.60 against $7 (Search, text and highlights of up to 10 results included). Contents · per 1k pages: $0.20 against $1 (Contents, text, per 1k pages). Contents with a query · per 1k pages: $0.40 against $2 (Contents, text + highlights, per 1k pages). No seats or minimums, and results served from the cache are free.

When should I stay with Exa?

If you need full page text on every result, written summaries, or people, company and research-paper search. Since the SDK is the same, you can also use both: send each request where it fits.

Switch from Exa

Free key, same SDK, 2.1⁠–⁠7× cheaper depending on the mode.

Get $5 freeTalk to sales