DeepSeek launches web search and page-fetching APIs for AI agents

DeepSeek's new APIs enable AI agents to access real-time web data, offering an alternative to reliance on static training data or limited internal datasets. The Search API delivers structured results, which can be directly parsed and utilized by agents for tasks requiring specific data points. The Fetch API, by converting web pages into Markdown, offers a clean format for agents to process content, potentially reducing the complexity of handling varied web structures and advertising.

The pricing model, RMB 5 per 1,000 successful requests, establishes a clear cost structure for developers integrating web search capabilities. This transparency allows for more predictable budgeting. Not charging for failed requests, empty search results, or timeouts reduces financial risk for developers, particularly during the testing and refinement phases of agent development.

Server-side search and fetching calls made through tools like DeepSeek Harness, Claude Code, and Codex are currently unbilled, with charges set to begin at midnight Beijing time on November 1. This grace period offers developers an opportunity to integrate and test the APIs without immediate cost. However, developers will need to account for these new costs in their operational budgets after the transition date.