DeepSeek Will Soon Charge for Its Web Search Tools

DeepSeek will begin per-request billing for its Search and Fetch APIs at 00:00 Beijing time on November 1, 2026. Each tool will cost $0.75 per 1,000 calls. Calls remain free until that cutoff. DeepSeek announcement DeepSeek pricing
The Search API finds web pages for a question and returns a structured list. The Fetch API opens one single page and returns its content as Markdown, a simple clean-text format. The two abilities are offered as separate tools. DeepSeek API Docs
Developers reach Search at POST https://api.deepseek.com/v1/tools/search. They reach Fetch at POST https://api.deepseek.com/v1/tools/fetch. Search is like asking a librarian for a list of books. Fetch is like copying one page to read.
Both tools use the same bearer authentication as the DeepSeek API, meaning the same secret key. No separate login system is needed. Existing key management, rotation, and proxy controls carry over directly.
Search runs on DeepSeek's own search engine. That engine handles hundreds of millions of requests daily worldwide, according to the company. Retrieval and generation come from the same provider, with speed and availability tied to one system rather than an outside search service.
Billing applies only when a request returns HTTP 200, the standard success signal, and produces a result. Failed calls, transport errors, and empty results are not billable. Clients can retry after a non-200 response or network timeout without paying for the failed try, though they will still pay for the eventual successful call.
The broader context here is predictable costs for AI systems that use fresh web facts. Per-call pricing separates web access cost from token accounting, or counting words processed. At $0.00075 per successful call, budgeting depends on how often the tools are called rather than page length or model context use. Teams running broad crawls, verification passes, or multi-hop research agents can estimate spending from logs of search and fetch calls alone.
In my view, keeping search and fetch separate gives clearer control over costs. Search calls stay cheap while testing questions, reordering results by relevance, or filtering websites on your own computer. Fetch costs happen only when a page is judged worth converting to Markdown. That setup rewards careful fetching and saving copies, and it penalizes fetching every search hit without choosing.
Looking at what this means for implementation, three details deserve attention. First, Markdown conversion moves parsing work off the client, but splitting text and adding citations must trust the provider's HTML-to-text conversion. Second, paying only for success simplifies accounting but requires careful logging on the client, since success status and presence of a result now decide spending. Third, shared key access lowers setup work, but it puts limits and misuse controls on a single key. Scope that key narrowly if search agents run in untrusted loops.


