DeepSeek Search & Fetch API: endpoints, $0.75/1k requests, free in Harness until Nov 1

On October 17, 2026, DeepSeek opened its in-house Search API for direct use. The Search & Fetch guide is the operational doc: two raw JSON endpoints on the same bearer key as chat, aimed at agents and RAG that need live web results without wrapping a third-party search vendor.

Builder question this answers: how do I call DeepSeek’s own web search and page fetch, what do they cost, and when does the free Harness path end?

Endpoints

Base URL: https://api.deepseek.com

API	Method	Path
Search	POST	/v1/tools/search
Fetch	POST	/v1/tools/fetch

Headers: Authorization: Bearer <DeepSeek API Key> and Content-Type: application/json.

Minimal Search call from the guide:

curl https://api.deepseek.com/v1/tools/search \
  -H "Authorization: Bearer <DeepSeek API Key>" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "DeepSeek V4 release",
    "count": 5,
    "need_content": true
  }'

Search body fields (official schema): query (required, 1–128 Unicode chars; supports site:, -site:, quoted phrases, lang:, after:, before:), count (1–50, default 10), optional ori_question, localization (IP for locale), need_content (default false). Each result includes rerank_score (0–3; ≥1.5 treated as high quality), title, url, snippet, summary, content, site_name, site_icon, date, and authority (0–4, or -1 unknown; ≥3 = highly authoritative).

Fetch takes a single url (HTTP/HTTPS, ≤8192 chars, no userinfo) and returns { "title", "content" } with Markdown body. (Source: Search & Fetch guide and Search API reference)

Pricing and the Nov 1 cliff

From Models & Pricing and the guide’s billing section:

API	Unit	Price
Search	Per 1,000 successful requests	$0.75
Fetch	Per 1,000 successful requests	$0.75

Billed per successful HTTP 200 that returns a result (Search needs a non-empty results array; Fetch needs page content). Empty searches, failures, and timeouts are not billed. count does not change the per-request price.

Important for agent users: when Search/Fetch run as server-side tools through DeepSeek Harness, Claude Code, or Codex, those calls are free until 00:00 Beijing time on November 1, 2026. After that date, the same per-request rates apply.

Rate limits and errors worth coding for

Per account, Search and Fetch each have a 5 QPS cap (independent of each other and of model concurrency). Over-limit requests return HTTP 429. Fetch also has a per-domain QPS cap. Other status codes called out in the guide: 400 bad request, 401 bad key, 402 insufficient balance, 403 Fetch blocked by security policy, 404 Fetch could not extract usable Markdown, plus 408/413/5xx/504 for timeouts and server faults. Retry only 408, 429, and 5xx; top up on 402.

How this fits a DeepSeek stack today

Use Search to shortlist URLs, then Fetch (or need_content: true on Search) for Markdown you can stuff into a RAG or agent context. Pair it with the same-day Flash-1017 model bump if your agent already routes through deepseek-flash. Setup for chat clients is still covered in the V4 Flash guide; peak/off-peak token rates for the model remain separate from these per-request Search/Fetch charges (peak pricing note).

Sources: Search & Fetch guide, Search API reference, Models & Pricing, Change Log 2026-10-17.