DeepSeek has expanded its API platform by introducing two new autonomous tools: a Search endpoint that queries the web and provides structured results, and a Fetch endpoint that extracts the content of a single page and returns it in Markdown format.
According to the Chinese API documentation, these tools are designed so that applications using agents and Retrieval-Augmented Generation (RAG) can access up-to-date internet content without needing to involve a third-party search provider.
Both tools are raw JSON APIs and use the same Bearer authentication as the rest of the DeepSeek API. Search is available at api.deepseek.com/v1/tools/search, and Fetch is available at api.deepseek.com/v1/tools/fetch.
Details of the Search function
The search request accepts a search query up to 128 characters long and can return between 1 and 50 results, with 10 set as the default. Developers can pass the original user question to help the service determine intent, specify an IP address for localized results, and request the full page text. Queries support operators such as site: and -site: to include or exclude domains, quotes for exact phrases, lang: to restrict languages, and after: and before: to restrict publication dates.
Each returned result includes a title, URL, a brief description of approximately 100 to 200 characters for display, a more detailed summary of about 600–800 characters intended for model reading, an optional full text field, a publication date, a re-ranking score from 0 to 3, and an authority rating from 0 to 4.
Data Retrieval Functionality
The Fetch tool accepts a single URL in HTTP or HTTPS format and returns the page header along with its content in Markdown format. DeepSeek notes that the results depend on the availability of the target site and its access policies, and blocked or unavailable pages will return an error.
Pricing Model and Limits
The cost is calculated per request, not per token or result. Both Search and Fetch cost 5 yuan per 1000 successful requests. Only calls that return HTTP 200 with content are billable: Search must return a non-empty list of results, and Fetch must return the page text. Failed requests, empty searches, and timeouts are free, and the number of requested results does not affect the price.
Requests initiated through the Web Search tool in Claude Code or the web_search function in Codex will not be charged for Search for the first two weeks after the Search API launch. DeepSeek states that standard pricing will apply afterward, and they will announce the exact billing start date separately.
Rate limits are set at the account level, not the key level. Search and Fetch are limited to 5 requests per second each, and these limits are tracked independently from each other and from model parallelism constraints. Fetch also has domain limits, and exceeding any of these limits results in an HTTP 429 response.