Shipped alongside the V4.1 Flash update, DeepSeek's standalone Search and Fetch APIs return ranked web results and clean page text as Markdown for $0.75 per 1,000 successful calls — free through its coding tools until November 1.
2026-10-20 · by Moe Ameen
On October 17, 2026, DeepSeek opened the search engine that powers its own app to outside developers, shipping two standalone HTTP APIs: Search and Fetch. Search takes a query and returns a ranked list of web results; Fetch takes a URL and returns that page's title and body as clean Markdown, stripped of navigation and ads. They landed the same day as DeepSeek's [V4.1 Flash -1017 checkpoint](/news/deepseek-v4-1-flash-1017), but they are not tied to it — the two tools are raw JSON endpoints (`/v1/tools/search` and `/v1/tools/fetch`) that authenticate with the same key as the rest of the [DeepSeek API](/ai-tools/deepseek-api) and can be called with no model involved.
Per DeepSeek's documentation, each Search result carries a title, URL, snippet, a longer model-oriented summary, optional full page content, publication date, site name, a reranking score that blends relevance, authority, and freshness, and a 0-to-4 source-authority rating. Both APIs cost $0.75 per 1,000 successful requests (RMB 5), and DeepSeek bills only HTTP 200 responses that return results or page content — failed requests, empty searches, and timeouts are not charged. Calls routed through DeepSeek's own coding tools (DeepSeek Harness, Claude Code, and Codex) are currently free; per-request billing for those begins at 00:00 Beijing time on November 1, 2026.
The move matters because it puts DeepSeek directly against grounding-and-retrieval vendors such as Tavily and Anthropic's native web-search tool. On published list prices, DeepSeek's $0.75-per-1,000-call rate works out to roughly 10x cheaper than Tavily's basic-tier search (about $8 per 1,000 calls) and roughly 13x cheaper than Anthropic's web-search tool ($10 per 1,000 searches) — though Anthropic's rate also bills the result tokens it feeds back to the model, a cost DeepSeek's flat per-call rate doesn't fold in, so the real-world gap narrows with heavy use. Treat the comparison as a list-price snapshot, not a benchmark of result quality — nobody has published a head-to-head on that yet.
A search-and-fetch API answers a different question than a chat model does: not "write this for me" but "what is out there, and read it cleanly." That is a research primitive — the raw mouth of a content pipeline. DeepSeek Fetch hands you a page as tidy Markdown; Search hands you ranked, dated, authority-scored results. Genuinely useful inputs. But a list of URLs and a wall of Markdown is not a post — it reframes nothing, films nothing, and reaches no audience.
[Kompozy](/) is the layer that turns a topic or a source into finished, scheduled content. It already has an input stage — RSS feeds, uploads, and webhooks flow into its pipeline — so a creator can hand it exactly what DeepSeek Search surfaced: a trending angle, a competitor's page, a fresh statistic. From that one seed Kompozy generates a [Persona Shorts](/glossary/persona-shorts) avatar video, a brand-exact [Carousel](/glossary/hyperframes), Quote Graphics, Photo Posts, a Blog Article, and an Email Newsletter, all held to one voice by the [Persona Brief](/glossary/persona-brief), then schedules and ships the set across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot) behind a per-post review gate. DeepSeek just made web retrieval cheap; Kompozy is what turns what you retrieve into published posts.
It is DeepSeek's in-house web search, opened to developers on October 17, 2026 as two standalone APIs: Search, which returns a ranked list of web results (with title, URL, snippet, a longer summary, publication date, and an authority score), and Fetch, which returns a given page's title and body as clean Markdown. Both are raw HTTP endpoints that use your DeepSeek API key.
Each costs $0.75 per 1,000 successful requests (RMB 5), and DeepSeek charges only for HTTP 200 responses that return results or content — failed requests, empty searches, and timeouts are free. Calls made through DeepSeek Harness, Claude Code, or Codex are currently unbilled; per-request billing for those starts at 00:00 Beijing time on November 1, 2026. Confirm current rates on DeepSeek's docs before budgeting.
No. They are standalone tool endpoints — you can call Search and Fetch directly, with any model or none, using the same API key as the rest of the DeepSeek API. That makes them usable as a general-purpose web-grounding layer, independent of DeepSeek's V4 models.
No. It finds and reads web pages and returns text — it generates no image, video, or audio and posts nothing. To turn what it surfaces into finished, scheduled content you pair it with a content engine like Kompozy, which generates the media and publishes across the eight social platforms plus blog and email.