According to official developer documentation published on api-docs.deepseek.com on October 18, 2026, DeepSeek has introduced native Search and Fetch API services designed to provide autonomous AI agents and retrieval-augmented generation (RAG) pipelines with real-time access to the public internet. The endpoints operate directly on DeepSeek's internal web indexing and crawler infrastructure, bypassing reliance on third-party commercial search aggregators.
The release divides functionality into two specialized developer endpoints. The Search API, hosted at /v1/tools/search, ingests user or agent queries and returns ranked collections of relevant web URLs, snippets, and domain metadata. Developers can toggle the need_content parameter to retrieve extracted body text directly within the search response, eliminating subsequent network round-trips. Complementing this, the Fetch API at /v1/tools/fetch accepts an arbitrary target URL, downloads the underlying HTML document, strips advertising scripts and boilerplate styling, and returns the sanitized content as clean Markdown formatted for large language model ingestion.
Pricing is anchored at $0.75 per 1,000 successful requests across both services, translating to approximately 5 RMB for international and domestic accounts. Crucially, DeepSeek established a success-only billing policy: applications are not billed for requests that encounter server timeouts, network interruptions, or queries that yield zero matching documents. Community analysis on Hacker News noted that this flat rate undercuts commercial web retrieval APIs such as Tavily and Exa, while avoiding the aggressive rate-limiting and CAPTCHA blocks associated with self-hosted scraping frameworks like SearXNG.
The introduction of first-party search tooling addresses a major operational bottleneck for autonomous coding and research agents. During iterative research runs, models frequently generate dozens of exploratory queries and web fetches; having low-latency, deterministic endpoints built into the model provider's cloud ecosystem curtails latency spikes and simplifies security credential management.
DeepSeek confirmed that while the Search and Fetch endpoints are accessible free of charge during initial ecosystem testing across tools such as DeepSeek Harness, Claude Code, and Codex, metered production billing will officially take effect on November 1, 2026, at 00:00 Beijing time (October 31 at 16:00 UTC). Integration documentation and SDK wrapper scripts are available immediately through the DeepSeek developer portal.