DeepSeek opened its search engine to developers through an API on October 17, 2026. Applications can now find web pages, inspect page authority ratings and retrieve content for AI workflows. The company also says its engine screens out search spam and attempts to manipulate generative systems.
What developers can connect to
DeepSeek says its engine indexes hundreds of billions of web pages and processes hundreds of millions of searches worldwide each day. The infrastructure powers search on its website and through its API, with support for multiple languages.
Two tools serve external applications. Search produces a structured set of results, while Fetch retrieves an individual page and returns its content as Markdown. Developers can use them together to locate a source and supply its text to a language model.
How pages are scored
The authority field assigns each page a tier from 0 to 4. A value of −1 means the rating is unknown, while tiers of 3 or higher identify sources DeepSeek considers highly authoritative.
A separate rerank_score runs from 0 to 3 and combines relevance, authority and freshness. DeepSeek considers results scoring at least 1.5 to be high quality. Its documentation does not disclose how these ratings are calculated.
Results include a short description for display and a longer summary for a model. Page content can also be requested when available. The site: operator restricts searches to a domain, with additional controls for language and publication dates.
Filtering claims and unanswered questions
DeepSeek says it filters SEO spam, low-quality AI-generated material and GEO manipulation. GEO refers to optimizing content to appear in answers from generative systems. The company has yet to explain which practices it classifies as manipulation or how detection works.
The role of outside search providers also remains unclear. DeepSeek’s privacy policy mentions third-party search APIs and sharing users’ keywords with those services. Bocha has described itself as an official web search supplier to DeepSeek, but the launch materials do not clarify whether that arrangement continues.
What site owners can learn
A single site: query can return up to 50 pages from a domain alongside their ratings. This offers a way to examine a site’s presence in the results. A missing page, however, does not establish that a filter was applied: the documentation provides no way to distinguish filtering from absence in the index or another cause.
The launch materials do not name the crawler responsible for the search index. The known DeepSeekBot is associated with collecting training data, leaving the effect of blocking it through robots.txt on search inclusion uncertain. Fetch access depends on the target website’s availability and access rules.
Pricing and the billing timetable
Search and Fetch each cost $0.75 per 1,000 successful requests. Searches with no results are not charged, and the number of pages returned within a request does not change its price.
For DeepSeek API users, server-side Search and Fetch calls through DeepSeek Harness, Claude Code and Codex are temporarily free. Billing for those calls is scheduled to begin on October 31, 2026, at 16:00 UTC.





Комментарии