Web search API pricing

What a web search API is

A web search API accepts a query and returns a provider-defined set of web results for discovery, retrieval, or grounding. The defining job is finding relevant web resources, not controlling a browser or fetching one already known URL. Products can return links, snippets, citations, scores, or retrieved content, and some can add generated answers or research modes. Those outputs are not interchangeable, so the table compares only the published standard-search meter. (Brave Search API, retrieved 2026-07-26; Exa API pricing, retrieved 2026-07-26; Perplexity Search API pricing, retrieved 2026-07-26.)

The closest neighbour is a SERP API. Both accept queries, but a SERP API is chosen when the buyer needs results from a named search engine with engine-specific parameters or result types. A web search API operates as its own search product and defines its own index, retrieval system, schema, or grounding contract. Brave Search API describes access to Brave’s search index, while SerpApi sells search requests as its plan unit. (Brave Search API, retrieved 2026-07-26; SerpApi pricing, retrieved 2026-07-26.)

A scraping API starts from a URL that the buyer already knows and returns that page or an extracted representation. A cloud browser provides a session that the buyer controls over time. AI extraction converts located or fetched content into a schema or answer. Search may lead into any of those steps, but a search-request price does not automatically include full page contents, browser interactions, or extraction tokens. Firecrawl, Browserbase, and Jina expose combinations of these jobs under separate or shared meters, which is why they can appear in more than one category. (Firecrawl pricing, retrieved 2026-07-26; Browserbase pricing, retrieved 2026-07-25; Jina AI Reader, retrieved 2026-07-25.)

What this category bills on

The direct meter is one standard search request. Some vendors charge per request, while credit products assign a credit weight to basic, advanced, deep, or research modes. Returned contents, answer generation, reranking, or follow-up extraction can add request, result, token, or page meters. Count the selected mode rather than assuming every query consumes the entry tier’s standard unit. (Tavily API credits, retrieved 2026-07-26; Linkup pricing, retrieved 2026-07-26; Perplexity Search API pricing, retrieved 2026-07-26.)

Per-result and token contracts need different billing units. ScrapeGraphAI records search consumption per returned result, so a call comparison would require an assumed result count. Jina’s public search record uses tokens but its current package price was not exposed in the reviewed public widget. A retired product also remains visible when it still defines the historical market but no longer has a current purchase path. (ScrapeGraphAI, retrieved 2026-07-26; Jina AI Reader, retrieved 2026-07-25; Microsoft Bing Search API retirement notice, retrieved 2026-07-26.)

The comparison workload

The table uses one standard web search request as the billable unit and shows every product carrying search-api in the dataset. It does not add generated-answer tokens, fetched-page contents, deep or research modes, reranking, browser time, or secondary extraction. When a source defines the unit as a request with a default result set, the calculation applies the meter once to the request rather than dividing it across results. Products with only a result, token, generic scrape, or retired meter remain in the table without an invented call price. (Scraping API Prices methodology, retrieved 2026-07-26.)

Rows are ordered by evidence state before bill size: assumption-free calculations, assumption-dependent calculations, published lower bounds, quote-required rows, and rows whose public meter does not fit the workload. Affiliate data does not affect inclusion or order. An assumption-marked row is a question to resolve with the vendor, not evidence for a cheapest claim. (Scraping API Prices methodology, retrieved 2026-07-26.)

What can change the bill

Query mode, result count, requested contents, citations, geography, freshness, depth, reranking, and answer generation can change the meter or add another meter. A workflow that searches and then fetches every result should budget the search call and the downstream page retrieval separately. A workflow that generates an answer should also check token or model charges instead of treating the plain search row as an all-in figure. (Exa API pricing, retrieved 2026-07-26; Tavily API credits, retrieved 2026-07-26; Perplexity Search API pricing, retrieved 2026-07-26.)

Products whose public meter does not produce this table’s bill

• Jina AI Reader and Search Foundation APIs provides Reader and Search products under token consumption. The reviewed public widget did not expose current package pricing, so the row requires a quote; Jina’s sales contact is the contract path. Retrieved 2026-07-25.
• ScrapeGraphAI combines search, scraping formats, structured extraction, crawling, monitoring, and optional stealth under shared credits. Search is recorded per returned result, so a call bill needs an assumed result count; its product page is the route to the entry tier or custom terms. Retrieved 2026-07-26.
• Olostep provides successful page retrieval with rendering and residential routing, plus a separately metered Answers workflow. The public record has no standard search-query meter, so its generic request rate is not reused; the product page is the route to the entry tier or custom scale terms. Retrieved 2026-07-25.
• Microsoft Bing Search API is retained as retired market inventory. Microsoft states that the Bing Search APIs were decommissioned and directs agent-grounding use cases to a different product, so there is no current raw Bing Search API contract path. Retrieved 2026-07-26.

How to use the comparison

Define whether the job needs links, snippets, citations, fetched contents, reranked results, or a generated answer before selecting a row. Price the standard search call first, then add result, page, token, browser, or extraction meters required by the workflow. Confirm the permitted use and returned schema separately. Treat not-computable as a procurement question and retired as unavailable for new raw-search purchases, not as missing data to fill with an estimate. (Perplexity Search API pricing, retrieved 2026-07-26; Microsoft Bing Search API retirement notice, retrieved 2026-07-26; Scraping API Prices methodology, retrieved 2026-07-26.)