DeepSeek launches Search and Fetch APIs for AI agent developers
DeepSeek has opened its in-house search engine to developers through new Search and Fetch APIs, priced at RMB 5 per 1,000 successful requests.
Chinese AI company DeepSeek has released two new developer services, Search and Fetch, aimed at teams building AI agents and applications that need to pull information from the web. Available as of October 19, 2026, the APIs give outside developers access to the company's proprietary search engine for the first time.
According to DeepSeek, the Search API returns web results in a structured format. The Fetch API takes a given webpage and extracts its title and body content as Markdown, a lightweight text format that language models can process easily.
Pricing and access
The company has set the same price for both services: RMB 5 per 1,000 successful requests. DeepSeek said it will not charge for failed requests, searches that return no results, or calls that time out. Developers can use their existing DeepSeek API credentials to reach the new endpoints, with no separate sign-up required.
DeepSeek also said that server-side search and fetch calls made through tools such as DeepSeek Harness, Claude Code and Codex are not being billed for now. Charges for that usage will start at midnight Beijing time on November 1, 2026, the company said.
Why it matters
AI agents acting on a user's behalf frequently need information that is newer than a model's training data. Web search and page reading have therefore become basic building blocks for agent-based software. By bundling these functions into its own platform, DeepSeek gives developers an option that does not require a separate search provider.
The mention of third-party coding tools is notable. Naming Claude Code and Codex alongside the company's own Harness suggests DeepSeek wants its retrieval infrastructure used across different development environments rather than only inside its own products.
Not billing for unsuccessful or empty requests may appeal to developers who want predictable costs when agents run large numbers of automated queries. Independent assessments of the services' speed and result quality have not yet been published.
Comments
No comments yet. Start the discussion or ask a question.
Related
Founded by two University of Lagos students, UniBeta abandoned a failed campus super-app and rebuilt itself as an AI tool that mines past papers for high-yield topics.
Apple reportedly plans to switch on Apple Intelligence in China with iOS 27.3 using Alibaba's Qwen models, while the new Siri, Image Playground and some writing tools stay off.
Amazon has introduced two personal AI gadgets that turn everyday conversations and phone calls into summaries, reminders and to-do lists.