DeepSeek prices its web search API at $0.75 per 1,000 calls

DeepSeek prices its web search API at $0.75 per 1,000 calls

17 October 2026: DeepSeek has opened two web tools to every API customer: a Search API that returns ranked results from its own search engine, and a Fetch API that returns a single page as Markdown. Each costs $0.75 per 1,000 successful requests. Teams whose coding agents run on DeepSeek through DeepSeek Harness, Claude Code or Codex are already calling both tools without paying for them, and that changes at 00:00 Beijing time on 1 November 2026, which is 16:00 UTC on 31 October.

Not yet independently verified. One outlet report gave the search price as 5 yuan per 1,000 calls; DeepSeek's own guide and pricing page state USD 0.75, which we use. The pricing and guide pages carry no date, and the price has not yet appeared in an independent report. We will update this when it can be confirmed, and remove this note.

Two endpoints behind the chat API key

The tools sit at /v1/tools/search and /v1/tools/fetch on api.deepseek.com and use the same bearer key as the chat models. A search request takes a query of 1 to 128 characters, asks for between 1 and 50 results (10 by default), and can pass the user's original question, an IP address to localise results, and a flag to return page content alongside each snippet. The query keeps its operators, so site:, -site:, exact phrases, lang: and after: or before: dates all work. Each result carries a title, URL, snippet, summary, date, an authority value and a rerank score from 0 to 3, where DeepSeek treats 1.5 or above as relevant.

In its change log entry of 17 October, DeepSeek says the engine already serves hundreds of millions of searches a day for its own web and API products, indexes hundreds of billions of pages, and filters SEO spam, manipulation aimed at AI answers, and low-quality machine-written pages.

How the bill is counted

Billing is per successful request, not per token and not per result. A call is billable only when it returns HTTP 200 and something useful: a non-empty results array for Search, page content for Fetch. Failed calls, empty searches and timeouts cost nothing, and asking for 50 results costs the same as asking for one. Charges come out of the account balance, with granted credit used before topped-up credit, and the pricing page reserves DeepSeek's right to change prices.

Rate limits are tighter than most teams will expect from a search provider: 5 requests a second for Search and 5 for Fetch, per account rather than per key, with Fetch also limited per target domain. Requests over either limit get HTTP 429. An agent that fans out several searches per step will hit that ceiling quickly, so retry logic matters more than usual.

The free calls that start costing money

The note buried in the guide is the one that affects existing users. Server-side tool calls made while DeepSeek runs inside DeepSeek Harness, Claude Code or Codex already go to these endpoints. Anyone who has pointed one of those agent tools at DeepSeek since the endpoints opened has been searching for free, and from 1 November every successful call adds to the bill. At $0.75 per 1,000, an agent making 2,000 searches a working day would add about $1.50 a day, small next to token costs but no longer zero, and it will show as a new line on invoices.

The model change that came with it

The same entry moved deepseek-flash to a new checkpoint, DeepSeek-V4.1-Flash-1017, with the model name and token prices unchanged. On the pricing page the peak rate is $0.30 per million input tokens on a cache miss, $0.006 on a cache hit and $1.20 per million output tokens, with off-peak at half those rates. The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work, but they are served by the new checkpoint and billed at the Flash price. We list DeepSeek V4.1 Flash and DeepSeek V4 Pro among the five DeepSeek models on our AI model board.

Sources

• DeepSeek API Docs: Change Log, entry of 17 October 2026, 17 October 2026
• DeepSeek API Docs: Search & Fetch guide (read 2026-10-18), 18 October 2026
• DeepSeek API Docs: Models & Pricing (read 2026-10-18), 18 October 2026
• MarkTechPost: DeepSeek releases V4.1-Flash-1017 and opens Search API, 17 October 2026