DeepSeek Search API Explained for Business Apps
DeepSeek opened its in-house search engine to developers on October 17, 2026. Learn how the Search and Fetch APIs work, what they cost, and how to build with them.
The Short Answer
On October 17, 2026, DeepSeek opened its in-house search engine to outside developers. The DeepSeek Search API lets any application query the live web and pull structured results back, so AI assistants, dashboards, and customer tools can answer questions with current information instead of relying only on what a model learned during training. It launched…
On October 17, 2026, DeepSeek opened its in-house search engine to outside developers. The DeepSeek Search API lets any application query the live web and pull structured results back, so AI assistants, dashboards, and customer tools can answer questions with current information instead of relying only on what a model learned during training. It launched alongside a refreshed model, DeepSeek-V4.1-Flash-1017, and a companion Fetch API that retrieves a single web page as Markdown. For businesses, the appeal is the combination of a simple two-endpoint design, pricing of $0.75 per 1,000 successful calls, and a free window for server-side calls through popular coding tools that runs until November 1, 2026.
What Is the DeepSeek Search API?
The DeepSeek Search API is a hosted web search service built for AI applications. You send it a query, and it returns ranked web results in structured form: titles, URLs, snippets, and metadata your code can use directly. It is powered by the same in-house search engine that already served DeepSeek's own website and app. DeepSeek says the engine indexes hundreds of billions of web pages, serves hundreds of millions of requests every day, supports multilingual search, and filters SEO spam, GEO manipulation, and low-quality AI-generated content. Those filtering claims have not been independently measured.
The practical difference from a general-purpose search API is the audience. This one is designed for programs, not people. Results come back in a format an agent or backend service can act on: pick the best sources, fetch the full pages, and cite or summarize them. That makes it a natural fit for retrieval-augmented generation (RAG), research agents, and support copilots.
The launch also included a Fetch API. Search finds pages. Fetch retrieves one specific page and returns its content as Markdown, which is far easier for a model to read than raw HTML. Together they cover the most common retrieval pattern in AI apps: search for candidates, then read the best ones in full.
How the DeepSeek Search API Works
Both endpoints live on DeepSeek's API host, api.deepseek.com, and accept POST requests authenticated with a DeepSeek API key as a bearer token.
| Endpoint | What it does | What it returns |
|---|---|---|
POST /v1/tools/search | Finds web pages for a query | Ranked, structured results with snippets and metadata |
POST /v1/tools/fetch | Retrieves one specific web page | The page content as Markdown |
Search takes a query of 1 to 128 characters. A count parameter controls how many results come back, from 1 to 50, with a default of 10. The number of results does not change the price: one search call costs the same whether it returns 5 results or 50. An optional need_content flag asks for page text inside the search response itself, which can save a separate Fetch call when a snippet is enough.
Search also supports operators that experienced search users will recognize: site: and -site: to include or exclude a domain, a leading minus to exclude a term, quoted phrases for exact matches, lang: for multilingual queries, and after: and before: for date ranges. Each result carries a rerank score from 0 to 3 and an authority tier from 0 to 4, which help your code decide which sources to fetch first. Fetch accepts a URL of up to 8,192 characters, which must use HTTP or HTTPS and may not carry an embedded username or password.
A typical pattern is: search with a tight query, rank results by rerank score and authority tier, fetch the top two or three pages, then pass the fetched Markdown to your model with clear citation instructions. A specialist partner can help wire that pipeline into an existing product, which is exactly what a good API development and integration service is for: auth, retries, caching, and monitoring handled as one coherent system.
DeepSeek Search API Pricing Explained
Pricing is simple. Each endpoint costs $0.75 per 1,000 successful requests. A request counts as billable only when it returns HTTP 200 with usable content: a non-empty results array for Search, or actual page content for Fetch. Failed requests, empty search results, and timeouts are not billed.
That billing rule matters for budgeting, because you only pay when you get something usable. It also makes a full search-and-fetch workflow easy to model. Each pair of one successful search plus one successful fetch costs $1.50 per 1,000 pairs. At 10,000 such pairs, the total is $15. At 10,000 successful requests to a single endpoint, the cost is $7.50, and at 100,000 it is $75.
| Monthly volume (successful requests) | Search only | Search plus Fetch pairs |
|---|---|---|
| 10,000 | $7.50 | $15.00 |
| 100,000 | $75.00 | $150.00 |
| 1,000,000 | $750.00 | $1,500.00 |
Two more details are worth noting. First, search and fetch are standalone retrieval endpoints. Model token usage is priced separately, so if you feed fetched pages into a model, you pay the retrieval fee plus the model's token charges. Second, there is a free window. Server-side calls made through DeepSeek Harness, Claude Code, and Codex are free until billing starts at 12:00 a.m. Beijing time on November 1, 2026 (4:00 p.m. UTC on October 31, 2026). Any production plan should assume full per-request billing from November onward.
Rate Limits and Fair Use
Each account is limited to 5 requests per second on Search and 5 requests per second on Fetch, counted separately. Using multiple API keys does not raise either limit, and Fetch also has a per-domain rate limit. For a prototype or a low-traffic internal tool, 5 requests per second is plenty. For a public-facing feature, it is a hard ceiling to design around. Practical ways to stay inside the limit include caching repeated queries, batching and debouncing so several users asking the same question share one search, queuing background work at a fixed pace, retrying with backoff when you do hit a limit, and fetching selectively using rerank scores and authority tiers.
Rate limits also shape product design. A feature that searches once per user request, caches the result, and refreshes in the background feels just as fresh as one that searches on every keystroke, at a fraction of the cost. Our guide to API integration in 2026 covers these patterns in more detail.
The Model Behind the Launch: DeepSeek-V4.1-Flash-1017
The Search API arrived with a refreshed model. DeepSeek-V4.1-Flash-1017 is a minor update to DeepSeek-V4.1-Flash that DeepSeek describes as delivering substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks, plus faster task completion and shorter outputs. It serves under the existing deepseek-flash API identifier, so existing integrations needed no changes.
The headline benchmark is DeepSWE v1.1, a software-engineering test, where DeepSeek reports 77.0, up from 74.2 for the previous Flash checkpoint. On Terminal-Bench 4.0, DeepSeek reports 42.3, up from 31.2. These are DeepSeek's own reported figures, not independently reproduced results, and benchmark scores are only one signal of real-world quality.
DeepSeek also posted the 1017 weights on Hugging Face under an MIT license, according to multiple launch reports. That removes the usual open-versus-closed tradeoff: use the hosted API for convenience, or self-host the weights where data control or latency demands it. Teams weighing that choice should read our overview of self-hosted AI models for business. If you want to combine hosted retrieval with a model strategy tailored to your product, our AI and machine learning services can help you design the full stack.
Business Use Cases for the DeepSeek Search API
Live web search is most valuable wherever answers go stale quickly:
- Customer support copilots. Agents and chatbots can search documentation, changelogs, and known-issue pages in real time, then answer with a cited source instead of a guess.
- Sales and competitor intelligence. A scheduled job that searches for a competitor's name plus an
after:date, then fetches the new pages, becomes a daily briefing with almost no manual effort. - Content and market monitoring. Marketing teams can monitor trending topics and coverage to inform content calendars. Date operators make it easy to ask what changed in the last 24 hours.
DeepSeek Search API vs Other Search APIs
Price is one of the most striking parts of this launch. At $0.75 per 1,000 successful requests, DeepSeek undercuts the listed prices of established search APIs by a wide margin. Brave lists its search API at $5 per 1,000 requests, and Exa's search starts at $4 per 1,000 requests, according to their pricing pages as of October 17, 2026. Those services differ in features, result quality, and coverage, so this is not a like-for-like comparison. On headline price alone, DeepSeek is roughly five to six times cheaper per successful call.
Price is not the only factor. Compare result quality by testing the same queries across providers, billing fairness, rate limits, latency from your own infrastructure, and data handling. For most small and mid-size teams, the sensible approach is to prototype on DeepSeek during the free window, measure result quality on your own queries, and only then commit.
Building With It: An Integration Checklist
- Secure the key. Keep the API key on the server, never in client-side code or a mobile app. Rotate it if it is ever exposed.
- Set timeouts and retries. Give each request a timeout, retry transient failures with backoff, and degrade gracefully when search is unavailable.
- Cache results. Cache search responses by normalized query and fetched pages by URL, with a time-to-live that matches how fast the topic changes.
- Validate and cite. Show users the sources behind AI-generated answers. Link back to the original pages and date-stamp retrieved content.
- Filter and moderate. Do not pass raw web content straight to users or into a model prompt without checks. Strip scripts and verify that fetched pages match the query.
- Log usage and plan for the limit. Track calls by feature and query type so you can see which features drive cost. Queue background jobs, debounce user-triggered searches, and build a fallback for the moment you hit 5 requests per second.
Limitations to Plan Around
- Throughput ceiling. 5 requests per second per endpoint is a hard cap. High-traffic consumer features will need caching, queuing, or a second provider as a fallback.
- Hosted dependency. The search engine is a hosted service. You cannot self-host the index even though the model weights are open, so your uptime depends on DeepSeek's infrastructure.
- Freshness varies. A search index updates continuously, but breaking news may lag. For time-critical alerts, verify freshness on your own queries.
- Results need verification. Search returns what the web says, including pages that are wrong or outdated. For decisions involving money, health, or legal matters, verify against primary sources.
- Pricing can change. The free window ends on November 1, 2026. Build your cost model so a price change is a configuration update, not a rewrite.
Frequently Asked Questions
What is the DeepSeek Search API?
It is a hosted web search service for developers, launched on October 17, 2026. You send a query to POST /v1/tools/search and receive ranked, structured web results that your application or AI agent can use. A companion Fetch API retrieves a single page as Markdown.
How much does the DeepSeek Search API cost?
Each endpoint costs $0.75 per 1,000 successful requests. You are only billed when a request returns HTTP 200 with usable content. Failed requests, empty search results, and timeouts are free. Server-side calls through DeepSeek Harness, Claude Code, and Codex are free until billing starts at 12:00 a.m. Beijing time on November 1, 2026.
What is the rate limit for the DeepSeek Search API?
Each account is limited to 5 requests per second on Search and 5 requests per second on Fetch, counted separately. Using multiple API keys does not increase the limit, and Fetch also has a per-domain rate limit. Caching, queuing, and selective fetching are the standard ways to stay inside the cap.
Can the DeepSeek Search API replace a general search engine?
For programmatic use, yes, in many cases. It is designed for applications and agents that need structured results, not for human browsing. For consumer-facing search products, compare result quality, coverage, latency, and data policies against established providers before switching.
Does the DeepSeek Search API work with other AI models?
Yes. Search and Fetch are standalone retrieval endpoints, independent of any particular model. You can feed their output into DeepSeek's models or into any other model.
Is the DeepSeek Search API good for business apps?
It is a strong option for apps that need current information: support copilots, research agents, competitor monitoring, and RAG systems. The low per-request price and free billing on failed calls make prototyping cheap. The main constraints are the 5 requests-per-second cap and the hosted dependency, so plan caching and fallbacks from day one.
Live web search turns an AI feature from a static answer machine into a current one. The DeepSeek Search API makes that capability cheap to try and simple to integrate, with two endpoints, fair billing, and a short free window to prove the value on your own data. The teams that benefit most will pair it with solid engineering: caching, rate-limit handling, and source verification. If you want help building a search-powered feature, a research agent, or a RAG system on this API, contact our team and we will scope it with you.
Related Articles
Keep reading — more guides you may find useful.
Related Services
Services connected to this guide.
