DeepSeek web Search and Fetch tool pricing for AI agents
DeepSeek ● Covered by 4 sources
DeepSeek priced its web Search and Fetch APIs at $0.75 per 1,000 successful requests each. The catch: failed calls, empty search results, and timeouts don’t get billed.
Based on reporting by DeepSeek — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
DeepSeek has put a price tag on the web tools its agents and RAG apps can use to reach live pages. Both Search and Fetch cost $0.75 per 1,000 successful requests. They’re billed separately, so looking things up and then pulling a page are two different meter ticks.
The Search API takes a query, returns structured results, and can include page content when asked. DeepSeek says it uses the company’s in-house search engine. The Fetch API does one job: take a URL, bring back the page, and turn it into Markdown. Both speak raw JSON and use the same bearer authentication as the main DeepSeek API.
The pricing model has a few sharp edges that matter to builders. A request only counts if it succeeds with HTTP 200 and produces usable output. Empty search results do not bill. Neither do failed requests or timeouts. And the bill is not tied to how many results you ask for, so the `count` field changes the output size, not the invoice.
DeepSeek is also drawing some clear lines around limits. Search can handle up to 100 requests per second per account, while Fetch is capped at 25. Those limits are account-wide, not per API key, and Fetch has an extra per-domain throttle on top. That means a busy crawler can hit two ceilings at once.
There’s one more wrinkle for people already using DeepSeek through server-side tools. If Search and Fetch are being called through DeepSeek Harness, Claude Code, or Codex, those calls are currently free. The billing clock starts at 00:00 Beijing time on November 1, 2026.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI pricing announcement that feels almost boring, and that’s a compliment. Flat per-successful-request pricing is easier to understand than token soup, and the free-to-fail rule keeps people from paying for the internet being the internet. The real test is whether the rate limits stay generous once everybody starts poking at live web data at once.
Read more about this at: DeepSeek
