Everything else in this directory is written for people choosing a search engine to use. This page is for people building something on top of one — retrieval for an AI agent, site search, or a research tool — and the market here has shifted more in the past year than almost any other corner of search.
Search APIs have quietly become one of the more volatile categories in this entire directory. Microsoft retired its long-standing Bing Search API outright in 2025. Google has closed free full-web access to its Custom Search JSON API to new customers and is pushing existing users toward a 2027 migration deadline. In their place, a newer generation of independent-index and AI-native providers — Brave, You.com, Perplexity — has built commercial APIs squarely aimed at the retrieval-augmented generation (RAG) use case: giving an AI system live, citable web content to ground its answers, rather than returning a results page for a human to click through.
The Bing Search API — long one of the two default choices, alongside Google's, for developers who wanted programmatic access to a general web index — was officially retired by Microsoft on August 11, 2025. All existing instances were decommissioned and new signups were closed prior to that date. Microsoft's designated replacement is Grounding with Bing Search, offered as part of Azure AI Agents rather than as a standalone consumer-facing search API; it's built specifically for grounding AI agent responses in current web content and requires provisioning through Azure rather than the old Bing API developer portal. Any documentation, tutorial, or code sample referencing the classic Bing Search API endpoints should now be treated as historical rather than actionable.
Google's Custom Search JSON API is not fully retired, but it is being wound down for the general full-web use case. As of early 2026, the API is closed to new customers for unrestricted full-web search; existing customers retain access but face a stated deadline of January 1, 2027 to transition to an alternative. The free tier remains capped at 100 queries per day, with additional volume priced at $5 per 1,000 queries up to 10,000 queries per day for those still grandfathered in. Google is directing developers toward two paths: Vertex AI Search (now positioned under the broader Agent Search umbrella) for building retrieval and grounding into an AI agent, or a full web-search solution available only by contacting Google directly, with pricing not publicly disclosed. Separately, the more narrowly scoped Custom Search Site Restricted JSON API — limited to searching up to 50 specified sites rather than the open web — continues to operate and remains the more realistic free option for site-search use cases going forward, though Google is steering that traffic toward Agent Search's searchLite method as well.
Brave offers programmatic access to its independently crawled index of more than 40 billion pages, positioned explicitly for both AI grounding and traditional search-enabled applications. It exposes separate endpoints for web, image, video, news, and suggestion/spellcheck queries, plus "Search Goggles" that let a developer re-rank or filter results according to custom rules rather than accepting Brave's default ranking. Pricing runs on a Search Plan ($5 per 1,000 requests, including $5 of free monthly credit and up to 50 queries per second) and a separate Answers Plan ($4 per 1,000 requests plus $5 per million tokens for AI-summarized answers), with custom Enterprise terms available for higher-volume or specialized needs.
You.com's Web Search API returns structured, LLM-ready results — web pages and news in a single call, with pre-extracted snippets, publication metadata, and optional full-page content via a "livecrawl" feature (2,000–10,000 words per result in HTML or Markdown). Pricing is $5 per 1,000 calls for standard web and news search, with livecrawl content retrieval priced separately at $1 per 1,000 pages; new accounts receive $100 in free credits. You.com also offers a separate Research API aimed at deeper, multi-step research tasks rather than single-query lookups.
Perplexity's developer offering, the Sonar API, is priced per token rather than per request: the base Sonar model runs $1 per million input tokens and $1 per million output tokens, Sonar Pro runs $3 input / $15 output per million tokens, and Sonar Reasoning Pro runs $2 input / $8 output per million tokens, each with an additional per-request search-context fee (roughly $5–$14 per 1,000 requests depending on how much search context is requested). A separate Sonar Deep Research tier adds citation and reasoning token pricing for multi-step research tasks. Because Perplexity's own consumer product dropped advertising entirely in February 2026 in favor of subscriptions, the Sonar API is worth evaluating specifically for the same reason: its incentives are aligned toward answer quality and citation accuracy rather than ad-supported placement.