DeepSeek Adds Search and Fetch APIs at 5 Yuan per 1,000 Calls
DeepSeek has added two standalone tools to its API platform, one that searches the web and one that fetches a single page. Both return raw JSON and use the same Bearer authentication as the rest of the DeepSeek API, so one key covers all three. The company says they let agent and retrieval augmented generation applications reach live web content without a separate search provider.
Search takes a query of up to 128 characters and returns between 1 and 50 results, with 10 as the default. According to Pandaily, developers can pass the user's original question to help the service read intent, supply an IP address to favour localised results, and ask for full page text. Queries accept operators to include or exclude domains, to require an exact phrase, to restrict language and to bound publication dates.
Each result carries a title, a web address, a short snippet of roughly 100 to 200 characters for display and a longer summary of roughly 600 to 800 characters meant for models to read. Results also carry an optional full text field, a publication date, a rerank score from 0 to 3 and an authority rating from 0 to 4. So a caller pays the same whether it asks for one result or fifty.
Fetch takes a single web address, over HTTP or HTTPS, and returns the page title plus a Markdown body. DeepSeek notes that results depend on the target site's availability and access policies. Blocked or unextractable pages return an error.
The two endpoints sit at api.deepseek.com/v1/tools/search and api.deepseek.com/v1/tools/fetch. Both take the same credentials, so a developer does not need a second vendor account. Between them, an agent can find sources and then read them inside one toolchain.
Pricing is per request rather than per token or per result, so cost tracks the number of calls rather than their size. Search and Fetch each cost 5 yuan per 1,000 successful requests. But only calls that return content are billed, so failed requests, empty searches and timeouts cost nothing.
DeepSeek is also waiving Search fees for use through the web search tool in Claude Code and the web search feature in Codex for the first two weeks after launch. Standard pricing applies afterwards, and the waiver covers Search rather than Fetch. The company says it will announce the exact billing start date separately.
Rate limits are set per account rather than per key, so extra keys do not buy extra throughput. Search and Fetch are each limited to 5 queries per second, tracked independently of each other and of model concurrency limits. Fetch also applies limits for each domain, and exceeding either cap returns an error code of 429.
For developers, the pitch is one bill and one set of keys for search, fetching and inference. But the tools are priced per call, so large scale crawling stays outside their scope. And DeepSeek's own note that blocked pages fail is a reminder that access policies still govern what an agent can read.
Search and Fetch are standalone endpoints on the DeepSeek API, returning JSON results and Markdown pages
Search returns 1 to 50 results with snippets, model readable summaries and an authority rating
Pricing is 5 yuan per 1,000 successful calls, with failed calls and empty searches free
Search and Fetch are each capped at 5 queries per second per account
Source: Pandaily
