DeepSeek
DeepSeek's In-House Search Engine Is Now an API: $0.75 per 1,000 Calls, Free Until November 1
DeepSeek quietly turned on two new endpoints this week — and then did something its competitors usually don't: it published the price on day one. The Search API and Fetch API are now live on the DeepSeek platform, powered by DeepSeek's own in-house search engine, priced at $0.75 per 1,000 successful requests each, and free until per-request billing starts at 00:00 Beijing time on November 1, 2026.
Verdict: If you're building RAG or agentic search on the DeepSeek API, start hitting these endpoints now — the free window is less than a week long, and there's no sign it will be extended. At $0.75/1K, Search lands at the cheap end of the commercial SERP API market, and it's the only search API that plugs straight into a frontier-lab model stack with the same API key. It's raw JSON (no chat-completion wrapper), so wiring it into an existing agent loop takes minutes, not days.
What actually shipped
Two endpoints, both raw JSON APIs using the same bearer authentication as the main DeepSeek API at https://api.deepseek.com:
- Search —
POST /v1/tools/search. Takes a query (1–128 characters), returns structured results with title, URL, and summary.countranges 1–50 (default 10),need_contentoptionally returns full page text, and alocalizationfield biases results by IP. Anori_questionfield passes the user's original question for intent disambiguation. - Fetch —
POST /v1/tools/fetch. Retrieves a single web page and returns its content as Markdown.
The search operators are Google-style: site:, -site:, exact phrases in quotes, exclusions with -, and lang:zh,en for language filtering. The interesting design detail: search syntax is passed through raw to the search service rather than parsed by the model layer, which keeps the endpoint predictable for programmatic use.
The kicker is in DeepSeek's pricing note: when you use DeepSeek through DeepSeek Harness, Claude Code, or Codex, server-side tool calls already hit the Search and Fetch APIs. Those calls are currently free — and per-request billing starts November 1. If you run agents through those harnesses, your usage is about to start showing up on the bill.
The pricing, and how it fits the market
Both endpoints bill identically at $0.75 per 1,000 successful requests, charged per request rather than per token. The word "successful" is doing real work there — failed requests presumably don't bill, which matters for flaky web targets where Fetch retries are common.
Where does $0.75/1K sit? Commercial SERP APIs cluster in the low-single-dollars-per-thousand range for similar structured results, with enterprise plans running higher; several offer free tiers that cap out at a few thousand queries. I'm deliberately not quoting per-competitor figures here — pricing pages move and I didn't re-verify each one today. What I'll say with conviction: DeepSeek entered the market at the aggressive end, consistent with how it has priced everything else. The bundled-model story is the sharper edge, though: one API key, one bill, one latency budget covering both the search and the reasoning over its results. That's a different buying decision than bolting a SERP API onto an OpenAI or Anthropic account.
The other structural fact: this is an in-house search engine, not a resold Bing or Google index. Microsoft retired its Bing Search APIs for new customers, and labs that want guaranteed search supply have been buying or building their own index ever since. DeepSeek building its own crawler-backed index is the same strategic move, at smaller scale — and it means result quality is the open question, not supply.
Result quality: the honest unknown
I haven't benchmarked result quality, and I won't pretend to have — this piece is a launch-and-pricing read, not a head-to-head eval. What the docs do establish: localization support, content inclusion on demand, operator support for site and language scoping. For English-language queries against fresh web content, expect a quality curve closer to a young search engine than to a two-decade incumbent. My advice for the free window: run your real RAG queries through it this week, compare against whatever you use now, and let the data decide before November 1 puts it on your bill. Free eval time with a fixed deadline is a genuinely good deal.
Who should care
- RAG pipeline builders on DeepSeek — obvious fit. Same auth, structured JSON, Markdown fetch output that drops straight into a context window.
- Agent developers using Claude Code or Codex against DeepSeek — you're already using this API without knowing it. Watch your usage dashboards after November 1; server-side tool calls start billing then.
- Cost-sensitive search buyers — $0.75/1K with no monthly minimum is worth a benchmark slot against your current SERP vendor.
- Anyone who assumed DeepSeek was chat-only — between Files API, context caching, and now search/fetch, the platform is quietly assembling a full agentic stack. We've covered DeepSeek's context caching economics before, and search slots neatly into the same play: cheap primitives, unbundled, pay-per-use.
One practical note for integrators: the legacy model names deepseek-v4-flash and deepseek-v4-flash-vision-exp are now retired — requests still work but are served by DeepSeek-V4.1-Flash-1017 at Flash pricing, per the current API docs. If your tooling still pins old model names, nothing breaks, but your cost model should reference the Flash rates.
Dated caveat: verified October 25, 2026 against DeepSeek's official API docs (endpoints, request fields, operator syntax, pricing, and the November 1 billing date pulled from api-docs.deepseek.com). Competitive SERP pricing is discussed qualitatively; per-vendor figures were not re-verified and are intentionally omitted. Accuracy is claimed only as of this date.