DeepSeek now sells web search and page fetch for $0.75 per 1,000 requests

Two plain JSON endpoints, one price, and a free ride until November 1 for anyone calling them through Claude Code or Codex. Commenters promptly lined up what Kagi, Brave and OpenAI charge for the same job.

In brief

DeepSeek published Search and Fetch APIs at $0.75 per 1,000 successful requests, billed only on HTTP 200 responses. Calls routed through Claude Code, Codex or DeepSeek's own harness are free until 00:00 Beijing time on November 1, 2026.

The first reply was a price check: Kagi's API runs $12 per 1,000 searches. Then a hobbyist who built his own agent tools said the big labs' bundled search costs 15-20x more than his.

Two endpoints, one price, and a billing date

DeepSeek social card
DeepSeek social card

DeepSeek's docs describe two raw JSON APIs under api.deepseek.com with bearer-token auth. POST /v1/tools/search returns structured results; POST /v1/tools/fetch returns one page as Markdown. Search takes a query of up to 128 characters with operators such as site:, lang:, after: and before:, and returns 1 to 50 results. Each result carries a rerank_score from 0 to 3 (1.5 and above counts as high quality) and an authority score from 0 to 4. Page content stays empty unless you set need_content.

Search Fetch
Price $0.75 per 1,000 successful requests $0.75 per 1,000 successful requests
Rate limit 5 QPS per account 5 QPS per account, plus a per-domain limit
Billed on HTTP 200 with results only HTTP 200 only

Empty searches, failures and timeouts cost nothing, and count does not change the price. Fetch can return 403 when a target blocks it by security policy. The sweetener: server-side calls made through DeepSeek Harness, Claude Code or Codex are free until 00:00 Beijing time on November 1, 2026.

The price comparison happened in the first five minutes

The thread went straight to benchmarks. A commenter with a personal site and GitHub link noted the same volume costs $5 with the Brave Search API and $12 with Kagi's. Another commenter added the Kagi detail that matters for agents: its $10-a-month plan is unlimited, but using those credentials with an LLM violates the terms.

"About a year ago I built my own search and fetch MCP tools for agents based on third-party scraping providers. It is about 15-20x more cost-effective than what OpenAI and Anthropic is offering with their integrated web search and fetch tools." — a developer who built his own agent search tools

One regular user, commenting on the model, not the API, said DeepSeek Flash is "neck and neck with SOTA models" and that search sets it apart. Another commenter predicted the cheap tier will not last:

"These prices will skyrocket eventually. AI companies need more money." — shevy-java

Who pays the sites being fetched?

The sharpest objections came from the other side of the request. A commenter running a large site said his company charges $1 per page crawled and that the only traffic getting through its firewall comes from hacked residential ISPs; he wanted to know whether DeepSeek's index contains his pages so he knows who to bill. Another asked whether DeepSeek will pay the creators whose pages are fetched. A third wanted to know how Fetch copes with Cloudflare's checkbox or Anubis; the docs only say success "depends on the target site's availability and access policies."

"These services only exist because Google refuses to provide a paid search API." — mjmvisser

For anyone running agents today, the practical read is the November 1 date: free-through-harness usage is a trial window, and the per-1,000 price in the table is what a metered agent will cost after it.

Get breakdowns like this every day

The top Hacker News stories, unpacked — by email, Telegram, or RSS.

Next story · 2 of 8 Read more
🔥Hot Discussion

A Theranos documentary 'deceives' its audience, and Hacker News disagrees about whether that's a review or a stunt

At a screening of the Elizabeth Holmes documentary, a woman yelled 'Criminal!' at the screen, and it wasn't aimed at Holmes. A Substack writer built a whole essay on that moment.

★ 368 · 💬 521 · readtheprofile.com · 4 min