DeepSeek Search & Fetch API pricing and features overview

DeepSeek Search & Fetch APIs cost $0.75 per 1,000 successful calls, making live web access cheap for LLM agents

DeepSeek announced two new raw‑JSON endpoints – Search and Fetch – that let language‑model agents retrieve live web results and page content. Both APIs are billed per successful request at $0.75 per 1,000 calls, regardless of the number of results returned.

Search API delivers structured results for a query

Key takeaway: The Search endpoint returns up to 50 ranked results with titles, URLs, snippets, summaries, and optional full‑page content.

  • Endpoint: POST https://api.deepseek.com/v1/tools/search
  • Authentication: Bearer token in the Authorization header; Content-Type: application/json required.
  • Request fields (all JSON):
    • query (string, 1‑128 chars, required)
    • count (int, default 10, 1‑50)
    • ori_question (string, optional, up to 128 chars)
    • localization (string, optional IP address)
    • need_content (bool, default false) – when true, the response includes the page’s full Markdown content.
  • Supported query operators include site:, -site:, -, -"...", "...", lang:, after:, and before:. Operators can be combined in a single query.
  • Response schema (200 OK):
    {
  "results": [
    {
      "rerank_score": 2.4,
      "title": "DeepSeek V4 model released",
      "url": "https://example.com/deepseek-v4",
      "snippet": "...",
      "summary": "...",
      "content": "...",   // present only if need_content=true
      "site_name": "Example News",
      "site_icon": "https://example.com/favicon.ico",
      "date": "2026-06-01T00:00:00+00:00",
      "authority": 3
    }
  ]
}
  • Scoring: rerank_score ranges 0‑3; scores ≥ 1.5 indicate high‑quality results. authority 3‑4 denotes highly authoritative sources.
  • Billing rule: A request is billable only when it returns HTTP 200 and the results array is non‑empty.
  • Rate limit: 5 QPS per account (independent of Fetch). Exceeding the limit yields HTTP 429.

Fetch API returns a page’s Markdown content

Key takeaway: The Fetch endpoint fetches a single URL and returns its title and full Markdown body, also billed at $0.75 per 1,000 successful calls.

  • Endpoint: POST https://api.deepseek.com/v1/tools/fetch
  • Authentication: Same bearer token scheme as Search.
  • Request field: url (string, required, max 8192 characters, must be a valid HTTP/HTTPS URL without credentials).
  • Response schema (200 OK):
    {
  "title": "Example article",
  "content": "# Example article\n\nArticle content."
}
  • Billing rule: A request is billable only when it returns HTTP 200 and the content field is non‑empty.
  • Rate limit: 5 QPS per account, plus an additional per‑domain QPS limit; violations return HTTP 429.

Error handling follows OpenAI‑compatible format

Key takeaway: Errors return a JSON object with error.message, error.type, and error.code; the param field is always null.

HTTP status Typical cause
400 Invalid JSON or field values (e.g., count out of range)
401 Missing or invalid API key
402 Insufficient account balance
403 Fetch blocked by target‑site security policy
404 URL unavailable, blocked, or non‑extractable content
408 Request body read timeout
413 Request body too large
429 QPS limit exceeded
500 Internal server error
504 Downstream timeout

Retry guidance: Wait and retry on 408, 429, or any 5xx error. Do not retry other 4xx errors without fixing the request.

Community perspective on pricing and limits

Key takeaway: Users compare DeepSeek’s $0.75/1k price favorably to competitors, but raise concerns about rate limits and anti‑bot robustness.

  • Pricing comparison: Brave Search API costs $5/1k, Kagi $12/1k, Exa $4/1k, while DeepSeek’s $0.75/1k is substantially cheaper.
  • Rate‑limit criticism: Several commenters note that 5 QPS may be insufficient for high‑traffic applications, especially when multiple concurrent users are expected.
  • Anti‑bot handling: Questions remain about how DeepSeek deals with Cloudflare challenges, CAPTCHA, or other bot protections; the documentation does not specify.
  • Alternative solutions: Some users point to free self‑hosted options like searxng‑ai‑kit or open‑source fetch services (Jina, Parallel, Firecrawl, Tavily, Exa) as more cost‑effective for personal projects.
  • Integration note: When using DeepSeek’s own Harness, Claude Code, or Codex, the Search/Fetch calls are currently free; billing starts at 00:00 Beijing time on 1 Nov 2026.

Practical usage tips

Key takeaway: To minimize cost and stay within limits, request only needed fields and reuse results where possible.

  1. Set need_content=false unless the downstream model explicitly requires full page text; this avoids unnecessary data transfer.
  2. Batch queries up to the count limit (max 50) to reduce the number of HTTP calls.
  3. Cache successful results for repeated queries; cached hits are not billed again.
  4. Respect QPS limits by adding exponential back‑off on HTTP 429 responses.
  5. Monitor billing via the DeepSeek dashboard; only successful, non‑empty responses count toward the $0.75/1k charge.

Bottom line

DeepSeek’s Search and Fetch APIs provide low‑cost, token‑agnostic web access for LLM agents at $0.75 per 1,000 successful requests, with a modest 5 QPS limit and OpenAI‑compatible error handling. While the price undercuts major competitors, developers must consider rate‑limit constraints, anti‑bot robustness, and the need to handle empty‑result billing nuances.

Sources

Related