DeepSeek has launched a pair of raw JSON APIs, 'Search' and 'Fetch,' to facilitate real-time web access for AI agents and RAG applications. The Search API utilizes DeepSeek's in-house search engine to identify web pages based on a user's query, returning structured results including titles, URLs, and summaries. The Fetch API complements this by retrieving the full content of a specific URL and converting it into Markdown format, a structure commonly used to maintain readability for LLM processing. Both services use the same bearer authentication as the standard DeepSeek API and are accessible via the `api.deepseek.com` base URL.

DeepSeek API documentation header for Search and Fetch services
DeepSeek's official documentation for the new Search and Fetch API offerings.api-docs.deepseek.com

This image represents the official documentation for the Search and Fetch APIs launched by DeepSeek.

The pricing model for these tools is based on successful requests rather than token consumption or the number of results returned. Both the Search and Fetch APIs are priced at $0.75 per 1,000 successful requests. A request is only billable if it returns an HTTP 200 status and produces a valid result—meaning the Search API must return a non-empty results array and the Fetch API must return page content. This structure ensures that users are not charged for failed requests, empty search results, or timeouts. Additionally, the Search API supports various query operators, such as `site:` to restrict results to specific domains, `lang:` for language filtering, and `after:` or `before:` for date-based constraints.

While the pricing is competitive, the service includes specific technical limitations and rate limits. Each account is subject to a 5 requests per second (QPS) limit for both Search and Fetch, which are tracked independently. For the Fetch API, there is an additional per-domain QPS limit to manage traffic to specific websites. The documentation also notes that Fetch requests are subject to the target site's availability and access policies, which may result in 404 errors if a site blocks the request or if the content cannot be extracted into Markdown. Furthermore, while current server-side tool calls through DeepSeek Harness, Claude Code, or Codex are free of charge, per-request billing for these tools is scheduled to begin on November 1, 2026.