Search API
A search capability API opened by DeepSeek that allows developers to integrate a retrieval-plus-generation combination into their own applications without building their own crawling, cleaning and ranking pipelines.
Core Facts
DeepSeek released the V4.1-Flash-1017 model on October 17, adding the Search and Fetch APIs.
DeepSeek launched a Search and Fetch API, split into two layers of capability: Search retrieval and Fetch web content extraction.
Timeline (last 90 days)
The Search API limit is 100 QPS per account, and the Fetch API limit is 25 QPS per account.
External search calls should set a timeout to avoid indefinitely hanging the worker process.
The example code requires the query string to be 1 to 128 characters long and the count parameter to be between 1 and 50, otherwise a ValueError is raised.
One search followed by fetching five pages is six API calls, not one, so cost should be calculated separately by number of calls.
Directly calling the two APIs Search and Fetch does not amount to having a research agent, because decision logic, source review, error recovery, and cost control are still missing.
Research agents and retrieval APIs are two different layers: a retrieval API is only responsible for fetching external information, whereas a research agent is a loop system that plans questions on its own, decides what to search for, which pages to select, and how to integrate the answers.
These two APIs use a billable-on-success model: a Search is billed only when it returns HTTP 200 with non-empty results, and a Fetch is billed only when it returns page content; empty results, failed requests and timeouts are not charged.
Search accepts a required query parameter, an optional result count count between 1 and 50, and a need_content parameter; Fetch accepts a URL and returns a title plus Markdown content.
The Search and Fetch APIs are both priced at USD 0.75 per 1,000 successful requests.
Search and Fetch use the same endpoint prefix; the endpoints given in the official guide are POST https://api.deepseek.com/v1/tools/search and POST https://api.deepseek.com/v1/tools/fetch, both using a Bearer API key and a JSON request body.
11 more timeline events
All Facts (20)
DeepSeek released the V4.1-Flash-1017 model on October 17, adding the Search and Fetch APIs.
65%Partially VerifiedDeepSeek launched a Search and Fetch API, split into two layers of capability: Search retrieval and Fetch web content extraction.
65%UnverifiedThe Search API limit is 100 QPS per account, and the Fetch API limit is 25 QPS per account.
50%UnverifiedOne search followed by fetching five pages is six API calls, not one, so cost should be calculated separately by number of calls.
50%UnverifiedDirectly calling the two APIs Search and Fetch does not amount to having a research agent, because decision logic, source review, error recovery, and cost control are still missing.
50%UnverifiedResearch agents and retrieval APIs are two different layers: a retrieval API is only responsible for fetching external information, whereas a research agent is a loop system that plans questions on its own, decides what to search for, which pages to select, and how to integrate the answers.
50%UnverifiedThese two APIs use a billable-on-success model: a Search is billed only when it returns HTTP 200 with non-empty results, and a Fetch is billed only when it returns page content; empty results, failed requests and timeouts are not charged.
50%UnverifiedSearch accepts a required query parameter, an optional result count count between 1 and 50, and a need_content parameter; Fetch accepts a URL and returns a title plus Markdown content.
50%UnverifiedSearch and Fetch use the same endpoint prefix; the endpoints given in the official guide are POST https://api.deepseek.com/v1/tools/search and POST https://api.deepseek.com/v1/tools/fetch, both using a Bearer API key and a JSON request body.
50%UnverifiedThe Search and Fetch APIs are both priced at USD 0.75 per 1,000 successful requests.
50%UnverifiedExternal search calls should set a timeout to avoid indefinitely hanging the worker process.
50%UnverifiedThe example code requires the query string to be 1 to 128 characters long and the count parameter to be between 1 and 50, otherwise a ValueError is raised.
50%UnverifiedThe Search API (POST /v1/tools/search) returns structured web results, including title, URL, summary, site authority tier, and relevance score, and is rate-limited to 5 requests per second.
50%UnverifiedThe DeepSeek search and scrape API is billed at $0.75 per 1,000 successful requests, i.e., $0.00075 per successful request, and the count parameter does not affect the price.
50%UnverifiedThe essence of model vendors opening Search APIs is packaging the engineering of "query rewriting—retrieval—reranking—stitching" into a service, so developers can get retrieval results with sources just by calling the interface.
50%UnverifiedThe risk of depending on a vendor's Search API is that retrieval quality, result freshness, and content compliance policies are all controlled upstream, so serious applications should design the retrieval layer as a replaceable adapter.
50%UnverifiedPerplexity opens its underlying retrieval capabilities to external developers through the Search API.
50%UnverifiedGitHub's search system is based on the Search API and supports a syntax system known as Search Qualifiers, including qualifiers such as is:open, comments:>, no:assignee, and org:.
50%UnverifiedDeepSeek opened its search API on the morning of October 17, priced at 5 yuan per 1000 calls, free until November 1.
50%UnverifiedDeepSeek has simultaneously opened up its Search API.
50%