orbita

WHAT YOUR AGENT GETS

The link
and the page’s text.

Your agent sends a query and gets up to 25 sources. Each comes with its address, its title, the publication date when the site gives one, and the page text in whole paragraphs, without menus or ads. No need to open and clean the pages yourself.

Over the API or a local MCP server.

API documentation ↗

QueryFast · 5 sources

In Pydantic AI's Amazon backend, which setting beats the unified one, and how do I get the reserved level?

  1. 01pydantic.dev/docs/ai/models/bedrock/

    AWS Bedrock

    Bedrock supports controlling the service tier to manage throughput and cost. You can use the unified service_tier field or the provider-specific bedrock_service_tier field. bedrock_service_tier takes precedence over the unified field when both are set.

    To request Bedrock’s 'reserved' tier (which requires a pre-purchased capacity reservation), set bedrock_service_tier directly — it isn’t reachable through the unified field.

  2. 02pydantic.dev/docs/ai/models/anthropic/

    Anthropic

  3. 03pydantic.dev/docs/ai/models/openai/

    OpenAI

How long
a search takes.

One search at a time on the 5.85-million-page index of the benchmark below, server time from the request to the response.

Fast

2.27s

medianp95 · 2.42 s

Standard

2.73s

medianp95 · 3.08 s

Deep

3.81s

medianp95 · 4.72 s

Measured on 24 October 2026 on a test server with the benchmark’s index: 60 questions per mode, the first 20 of SimpleQA, Natural Questions and TechQA-open. In the benchmark itself, 16 searches at once on one machine, the median was 2.68 s in Fast, 3.61 s in Standard and 5.03 s in Deep. Network time to the client is extra.

7 QUESTION SETS · 24 OCTOBER 2026

Orbita next to
Exa and Linkup.

The same 1,412 questions from seven sets for every service, and one check: is a correct answer in the text it returns. Each searched its own index: Exa and Linkup the whole web, Orbita a test index of 5.85 million pages built for this run (English Wikipedia and 1,927 other sites), not its smaller production index. On technical documentation, where Orbita has the pages, it is close to them or ahead; on general knowledge many answers are on pages it does not have.

    Every question and the method ↗

    The figures behind the charts. 24 October 2026: the share of questions whose answer appears in the text each service returned. Orbita Fast returns up to 5 results, Standard 10, Deep 25; Exa Auto and Linkup Standard 10. Exa and Linkup searched the whole web; Orbita its test index of 5.85 million pages, built for this run.
    MeasureOrbita FastOrbita StandardOrbita DeepExa AutoLinkup Standard
    SimpleQA: answer in the returned text, % of 30060.363.066.090.086.7
    Natural Questions: answer in the returned text, % of 30076.078.783.089.087.3
    HotpotQA: answer in the returned text, % of 25059.264.472.474.862.0
    AA-Omniscience: answer in the returned text, % of 20039.548.054.581.570.5
    BrowseComp: answer in the returned text, % of 1005.09.013.027.011.0
    TechQA: answer in the returned text, % of 14495.195.896.591.092.4
    TechQA-open: answer in the returned text, % of 11893.295.896.6100.097.5

    PRICING

    What search
    and Fetch cost.

    01

    Fast

    1–5results

    Quick facts, weather, rates and one command from the docs.

    $1.99New Search · per 1 000
    $2.99Search + Compact

    02

    Standard

    6–10results

    Comparisons, error fixes, harder docs and this week’s news on a topic.

    $3.99New Search · per 1 000
    $5.99Search + Compact

    03

    Deep

    11–25results

    A person, a company, a timeline or news across regions.

    $7.99New Search · per 1 000
    $12.99Search + Compact

    04

    Auto

    Autorequested
    →
    Fast–Deepresolved mode

    Orbita picks Fast, Standard or Deep for each query, within your limits.

    $1.99–$7.99resolved Search price
    $2.99–$12.99resolved Compact price
    Never exceeds max_mode or max_cost_usd

    05

    Fetch

    Any page, or a site section of up to 100 pages per request, as clean text in whole paragraphs.

    • Text from images (OCR)
    • JavaScript when needed
    • Every PDF page
    • Failed pages are free
    $0.50per 1,000 pagesExa · Parallel · You.com $1

    Fetch next to the others

    The whole page as clean text in whole paragraphs, text from images and every PDF page, at one price. The others return markdown.

    Service · returnsPrice per 1,000 pagesText from imagesFailed pages
    Without JavaScriptWith JavaScriptSite crawl
    Orbita Fetchwhole page: clean text in whole paragraphs$0.50$0.50$0.50IncludedFree
    Exa Contentsmarkdown$1.00$1.00Subpages, price not publishedNot stated?
    Parallel Extractmarkdown$1.00$1.00NoNot stated?
    You.com Contentsmarkdown, html$1.00$1.00NoNot stated?
    Linkup Fetchmarkdown$1.00$5.00NoNot statedFree
    Tavily Extractmarkdown, text$1.00–1.60$2.00–3.20$1.50–2.40NoFree
    Firecrawlmarkdown, html, json$0.60–3.80$0.60–3.80$0.60–3.80Not stated403 and 404 are billed
    Spidermarkdown, text~$0.02–0.10~$0.04–0.16~$0.30Not statedFree
    Scrapflymarkdown, text$0.09–0.15$0.55–0.90$0.09–0.15Not statedFree

    USD per 1,000 pages, from each service's pricing pages on 21 October 2026. A range means the price depends on the plan: from the largest self-serve plan to pay as you go; for Spider, the averages from its own guide. ? means not published.

    Every mode returns sources with links and publication dates. Compact adds a controlled context pass.

    • New Search$1.99–$7.99
    • Search + Compact$2.99–$12.99
    • Linkup$5
    • Tavily$5–8
    • Exa$7
    USD per 1,000 successful searches. Failed requests are refunded. No paid checkout is connected yet.

    OUR PRINCIPLE

    The web can be endless.
    Search should not feel chaotic.