DeepSeek opened its in-house Search and Fetch interfaces to developers on October 17, giving external applications access to web search and content extraction capabilities that were previously internal to the company's own products.
With the two interfaces, DeepSeek models can obtain external information, read web pages, and combine those abilities with code execution and file processing to complete continuous multi-step tasks. The company said this extends use cases from offline question answering to connected research, data analysis and engineering development.
Both interfaces are priced at 0.75 dollars per 1,000 successful requests. Server-side calls made through DeepSeek Harness, Claude Code and Codex are temporarily free, with billing set to begin on November 1, Beijing time.
DeepSeek also optimized its model and tooling for mainstream coding and office agent frameworks and released model weights under the MIT license, making it easier for developers to reuse existing tool chains, deploy privately and customize for specific business scenarios.
Analysts say the move lowers the cost of building agents on top of DeepSeek and keeps the company competitive in the intensifying race to supply the tooling layer, not just the model layer, of the AI stack. It follows the company's October 17 update of its V4.1 Flash model, which focused on full-project development capability and shorter task runtimes.
Source: China Foundation News / Phoenix Tech citing company announcements, October 17-20, 2026.
