DeepSeek opens standalone Search and Fetch APIs at $0.75 per 1,000 requests

DeepSeek now sells web Search and Fetch-as-Markdown endpoints on its regular API key, billed per successful request.

DeepSeek has opened two raw JSON endpoints, Search and Fetch, that give agents and RAG apps live web access on the same bearer key as its models, according to its Search & Fetch guide. Search returns ranked results with snippets, model-oriented summaries, optional page content, a rerank_score and an authority tier. Fetch returns one page as Markdown. Both cost $0.75 per 1,000 successful requests. Empty results, failures and timeouts aren't billed, and count (up to 50) doesn't change the price. The New Stack notes that neither endpoint requires a DeepSeek model call, so you can feed the results to any model.

The guide lists account-level limits of 100 QPS for Search and 25 for Fetch, plus a per-domain limit on Fetch. Several launch-day write-ups, including TokenCost, cited 5 QPS, so check the docs before you size a worker pool. Server-side tool calls through DeepSeek Harness, Claude Code or Codex already hit these APIs and are free until 00:00 Beijing time on November 1, 2026.

On price, TokenCost says OpenAI and Anthropic charge $10 per 1,000 built-in searches. It puts Exa's page contents at $1 per 1,000 and Perplexity's Agent API fetch_url at $0.50. THE D[AI]LY BRIEF lists Tavily basic search at $8.

The caveats are real. The same publication says DeepSeek's terms are governed by PRC law and that the search guide says nothing about retention. In one hands-on test, authority behaved like a per-host score. Fetch also identified itself as Firefox from Huawei Cloud in Hong Kong and returned pages that robots.txt disallows. dev.to adds a cost warning: 50 summaries add 30,000–40,000 characters to your context, and you pay for those as model tokens.

Sources