DeepSeek just made a quiet move that could matter more than another model release: it is selling access to its own search engine. The company launched Search and Fetch APIs for developers building agents and apps that need outside information.
The split is simple. Search returns structured web results. Fetch takes a URL and returns the page title and content as Markdown, which is the format most LLM pipelines want anyway. Both cost RMB 5 per 1,000 successful requests, roughly 70 US cents. Failed requests, empty results and timeouts are free. That last detail is a good one. Plenty of search APIs charge you for calls that return nothing useful.
Access runs through existing DeepSeek API credentials, so there is no new account, billing setup or key to manage. For anyone already using DeepSeek models, that removes real friction.
There is also a billing wrinkle. Server-side search and fetch calls made through tools like DeepSeek Harness, Claude Code and Codex are unbilled for now. Charging starts at midnight Beijing time on November 1. So the current free window is basically a trial period, and developers should expect their costs to change in two weeks.
Why does this matter? Because search has become the unglamorous bottleneck of agent building. A model is only as useful as the data it can reach, and most teams stitch together a search provider, a scraper and a cleanup step. Tavily, Exa, Brave’s Search API and Perplexity’s Sonar API all sell versions of this. Google and OpenAI bundle search into their own platforms. DeepSeek is now offering the search and fetch pieces as a clean pair, attached to a model provider developers already like for low prices.
The mention of Claude Code and Codex is the most telling part. DeepSeek is not only courting people who use its models. It wants to be the retrieval layer even inside rival coding agents. That fits its recent push to open-source Harness to developers, which came as a counter to Anthropic’s Claude Cowork. The strategy looks like owning more of the agent stack, not just the model at the bottom of it.
Still, some caveats. Pricing is quoted in RMB, and the announcement was published in Chinese. We don’t know how index freshness, coverage of English-language sites, rate limits or latency compare with Western rivals. A cheap API with weak results is not a bargain. And teams with data residency concerns will want to look closely at where queries go.
My read: this isn’t noise. If the quality holds up, DeepSeek has set a low price anchor for agent search, and competitors will have to answer it. Test it before November 1, while it’s free.
