Evergreen Tools
|
← Back to AI Tools

DeepSeek Search & Fetch API: Web Retrieval for Agents

Every agent eventually needs to look things up on the web. DeepSeek opened its own search stack as two standalone endpoints: Search returns results, Fetch pulls a page back for the model to read. The official pricing page lists $0.75 per 1,000 successful requests for each, and server-side calls made through DeepSeek Harness, Claude Code or Codex start billing at 00:00 Beijing time on 1 November 2026

Tool Interface

Interactive tool will be available soon

Features

  • ✓ Billed per successful request at $0.75 per 1,000, with Search and Fetch counted separately
  • ✓ Per the official wording, only successful requests are charged, so failures, empty results and timeouts do not count
  • ✓ Powered by DeepSeek's in-house search engine, which the company says serves hundreds of millions of requests a day
  • ✓ Kept as a separate line item from model tokens, so your cost sheet has two columns rather than one bundled price
  • ✓ Server-side tool calls inside DeepSeek Harness, Claude Code and Codex are free today and bill at the same rate from 1 November 2026

How to Use

  1. Confirm the current price and effective date on the official page: https://api-docs.deepseek.com/quick_start/pricing
  2. Budget at $0.00075 per call and remember one search plus one fetch is two billable requests, not one
  3. Audit existing agent loops: anything that grounds itself in live search through Harness, Claude Code or Codex starts costing real money after 1 November
  4. Add caching and de-duplication, because re-fetching the same URL burns budget for nothing

FAQ

How much does it cost and how is it counted?

The official pricing page lists $0.75 per 1,000 successful requests for Search and the same for Fetch, billed per request rather than per token. Source: https://api-docs.deepseek.com/quick_start/pricing

When does billing start?

The pricing page states that server-side Search and Fetch calls made when you use the DeepSeek API through DeepSeek Harness, Claude Code or Codex are currently free, with per-request billing starting at 00:00 Beijing time on 1 November 2026. Direct API calls already bill at the listed rate.

Do failed requests cost anything?

The price is defined per successful request, which is exactly how the table's unit reads. For the exact treatment of failures, empty results and timeouts, check the billing rules in the official Search & Fetch guide.

Is this the same as model token cost?

No. The pricing page keeps them apart: models are priced per million tokens (deepseek-flash is $0.15 per million input on a cache miss and $0.6 per million output at off-peak rates), while retrieval is priced per request.

Is it cheaper than OpenAI's or Anthropic's built-in search?

Third-party round-ups put OpenAI's and Anthropic's built-in web search tools at around $10 per 1,000 searches before tokens, against $0.75 per 1,000 here. That comparison is secondary data, so verify each vendor's own page before you sign anything.