DeepSeek launches search API: $0.75 per 1,000 requests

DeepSeek has opened access to two web tools aimed at agents and RAG applications that need live access to web content. Search finds pages for a query and returns structured results; Fetch retrieves a single page and hands back its content as Markdown. The search runs on the company's own engine.

Both are plain JSON APIs that use the same bearer-token authentication as the main DeepSeek API. Requests go to api.deepseek.com, with separate endpoints for search and fetch.

What search can do

The query supports a familiar operator syntax: restricting to or excluding sites (site:), filtering out individual terms and exact phrases, and limiting by language (lang:) or publication date (after:, before:). The count parameter sets the number of results, from 1 to 50.

Each result comes not only with a title, URL and short snippet, but also a longer 600–800-character summary meant to be read by a model rather than a human. There are two quality signals too: a rerank_score from 0 to 3 (blending relevance, authority and freshness, with 1.5 and above deemed high-quality) and an authority tier from 0 to 4.

Pricing

Search and Fetch cost the same: $0.75 per 1,000 successful requests. Billing is per request, not per token or per returned result — count does not affect the price. A request is billable only when it returns HTTP 200 with a non-empty result. Failed calls, empty search results and timeouts are not charged.

One detail for those already using DeepSeek through DeepSeek Harness, Claude Code or Codex: server-side tool calls there already hit these APIs. They are free for now, but per-request billing kicks in at midnight Beijing time on November 1, 2026.

Limits

The rate limit is 5 requests per second for each tool, counted independently and at the account level rather than per key. Fetch carries an extra per-domain limit on top of that. Both tools are read-only; on 429 and 5xx errors, the docs suggest waiting and retrying.

Source: Hacker News