NewYour Exa or Tako SDK, up to 5× cheaper.

Compared with Parallel

vsParallel

The Parallel alternative for news search

Parallel searches the whole web for agents, with compact excerpts and a cheap fast mode. typesearch is built for the news: every result with its outlet, publication time and a calibrated relevance probability, the top ones read before they’re ranked, tone and same-story grouping included.

Checked Sep 25, 2026 · A web search API for agents, with monitors

Mode by mode, per 1,000 searches

  • typesearch · Fast

    $1.40per 1,000

    Parallel: $129% less on Parallel

    Search API, fast mode

  • typesearch · Normal

    $2.20per 1,000

    Parallel: $5−56%

    Search API, basic mode

  • typesearch · Deep

    $5.60per 1,000

    Parallel: $511% less on Parallel

    Search API, advanced mode

Parallel’s Search API is $1 per 1,000 in turbo and fast modes and $5 in basic and advanced, for 10 results; more results cost extra. Its fast mode costs less than ours; its advanced mode costs a little less than our deep, which reads 8 articles with highlights.

Checked Sep 25, 2026 · Sources: parallel.ai/pricing

Side by side

What Parallel publishes, next to what typesearch does. Each of their facts links to where we read it.

Side by sideParalleltypesearch
PricingPer request: Search $1 (turbo, fast) or $5 (basic, advanced) per 1,000; Task, Monitor and more Pay as you go, per search: $1.40 fast, $2.20 normal, $5.60 deep, per 1,000. No plan, no minimum
Free planUp to 5,000 requests a month free; up to $80 at signup plus $5 a month $5 in credit once, valid 30 days — not limited to development or testing
How far backThe web, filtered with after_date A rolling 30-day window: built for what’s happening now
CoverageThe web; a country (location) parameter, no language parameter Sources in 130+ countries and 30+ languages; counts by country and language in GET /v1/sources
Full article textExcerpts in search; full pages with Extract No: title, the outlet’s standfirst and short verbatim excerpts, always with the link
Per request10 by default, up to 20 Up to 50 judged results; up to five queries per request
RankingRanked for a natural-language objective; publish_date on results A calibrated relevance probability per result; the top ones read and verified in normal and deep
EnrichmentCompact excerpts for LLMs; Monitor with webhooks as a separate product Your own typed questions per result, tone, same-story grouping, who published first
Rate limit600 searches/minute 10 requests/second; 50 with a monthly plan
SDKsPython and TypeScript OpenAPI 3.1; works with the official Exa, Tako and Perplexity SDKs
MCP serverYes, remote, free without a key Yes, remote: api.typesearch.ai/mcp

Checked Sep 25, 2026 · Sources: parallel.ai/pricing, docs.parallel.ai/api-reference/search/search.md, docs.parallel.ai/search/advanced-search-settings.md, docs.parallel.ai/extract/extract-quickstart

How the parameters map

Each Parallel parameter and what it becomes on typesearch. You rewrite one request; your agent’s logic stays.

Parallel parameterStatusOn typesearch
objective, search_queriesMappedquery: one, or up to five queries, judged together.
mode: turbo, fast, basic, advancedMappedmode: "ultra", "fast", "normal", "deep"
source_policy.include_domains, exclude_domainsSameinclude_domains, exclude_domains, up to 20 each.
source_policy.after_dateMappedpublished_after, and published_before for the other end.
locationLimitedQuery in the language you want, or restrict to regional outlets with include_domains.
max_resultsSamemax_results, up to 50.
excerptsMappedhighlights: verbatim excerpts from the articles that were read (normal and deep).
fetch_policy.max_age_secondsLimitedfresh: true skips our 10-minute result cache.
x-api-keyMappedAuthorization: Bearer <key>
results[]: url, title, publish_date, excerptsMappedresults[]: url, title, published_at, highlights — plus source, snippet and score

What you gain

  • The outlet and its standfirst on every result, not just a URL
  • A calibrated relevance probability you can threshold
  • Tone, same-story grouping and typed answers in the same call
  • Date windows on both ends (Parallel filters by after_date only)
  • Normal mode reads and verifies the top 4 for $2.20 per 1,000, against $5 for Parallel basic

When Parallel is the better fit

We’d rather you pick right the first time. Keep Parallel for these:

  • You search the whole web, not only news.
  • You need the cheapest fast search: Parallel fast is $1 per 1,000; ours is $1.40.
  • You need monitors with webhooks today (ours are coming).

The migration

The whole migration

Before and after: the request, the key header and the fields you read. Everything else in your code stays.

import requests
 
-r = requests.post("https://api.parallel.ai/v1/search",
- headers={"x-api-key": PARALLEL_API_KEY},
- json={"search_queries": ["lithium royalties in Chile"], "mode": "fast",
- "advanced_settings": {"source_policy": {"after_date": "2026-09-18"}}},
-).json()
+r = requests.post("https://api.typesearch.ai/v1/search",
+ headers={"Authorization": f"Bearer {TYPESEARCH_API_KEY}"},
+ json={"query": "lithium royalties in Chile", "days": 7, "max_results": 20, "mode": "fast"},
+).json()
 
-for a in r["results"]:
- print(a["title"], a["url"], a["publish_date"])
+for a in r["results"]:
+ print(a["title"], a["url"], a["published_at"], a["source"], a["score"])
beforeafter

Switching from Parallel

How is typesearch different from Parallel?

Parallel ranks the web for a natural-language objective and returns compact excerpts. typesearch ranks the news for your query with a calibrated probability, reads the top results in normal and deep, and adds tone, same-story grouping and typed answers.

How much does typesearch cost compared with Parallel?

Fast: $1.40 per 1,000 on typesearch, against $1 (Search API, fast mode). Normal: $2.20 per 1,000 on typesearch, against $5 (Search API, basic mode). Deep: $5.60 per 1,000 on typesearch, against $5 (Search API, advanced mode). Results served from the cache are free, and there are no seats or minimums.

How do I switch from Parallel?

Rewrite one request: the endpoint, the key header and a few parameter names change — see the table above. The fields you read map one to one: publish_date → published_at, excerpts → highlights (plus source and snippet, which Parallel doesn’t return).

When should I stay with Parallel?

If you need the whole web beyond news, the cheapest fast search, or monitors with webhooks today. You can also use both: typesearch for what your agent searches now, Parallel for the rest.

Switch from Parallel

A key in a minute, $5 in free credit, one request to rewrite.