The DeepSeek Search API lets your app or AI agent search the web and pull page text for $0.75 per 1,000 successful calls. That is far below most rivals, but it comes with a low rate limit and servers in China. DeepSeek announced it on October 17, 2026, alongside its new V4.1 Flash model.
Here are the quick facts:
- Two endpoints: Search finds pages, Fetch returns a page as clean markdown.
- Price: $0.75 per 1,000 successful requests for each endpoint.
- Free through DeepSeek Harness, Claude Code and Codex server-side tools until October 31, 2026 at 16:00 UTC.
- Open to everyone, according to DeepSeek’s announcement.
The low price is the headline. The catches below decide whether it fits your project.
How the DeepSeek Search API and Fetch API work
There are two separate tools. Search runs a query on DeepSeek’s own search engine and returns relevant results. Fetch takes a URL and returns the main text of that page in markdown, which is easy for a language model to read.
Most people will chain them: search first, then fetch the best hits for summarizing or research. DeepSeek says its in-house engine already serves hundreds of millions of requests a day, per its announcement.
If you want the model side of this launch, our DeepSeek V4.1 Flash 1017 price and specs post covers it.
DeepSeek Search API pricing next to Brave, Tavily and Exa
A developer write-up on DEV Community compared list prices. The figures below are that author’s own estimates for 10,000 agent runs, each using 5 searches and 3 fetches. I could not check every provider’s pricing page, so treat them as a rough guide.
| Provider | Price per 1,000 searches | Est. cost, 10,000 runs |
|---|---|---|
| DeepSeek (search + fetch) | $0.75 | $60 |
| Exa (instant) | $4 | $230 |
| Brave (search only) | $5 | $245 |
| Tavily (basic, pay as you go) | $8 | $448 |
| Gemini grounding | $14 | $630 |
The gap is large enough to matter for batch jobs and testing. It matters less if you run a handful of searches a day.
The catches to check before you switch
Cheap does not mean drop-in. According to the same DEV Community review, these are the limits to plan around:
- Rate limit: reportedly 5 queries per second per account for each endpoint, and extra API keys do not raise it.
- Query length: queries are capped at 128 characters, so long agent-written prompts need trimming.
- Data location: DeepSeek’s privacy policy puts its servers in China, so every query and URL goes there.
- Billing: only responses with content are charged, but an irrelevant result still counts as a success.
The rate limit is the one most likely to bite. A busy app with many users searching at once would hit errors quickly.
Quality is also unproven. DeepSeek’s announcement does not publish rate limits, and no public benchmark against Exa or Tavily exists yet. A weak result that forces extra retries can wipe out the savings.
Free through Claude Code and Codex, for now
If you use Claude Code or Codex with DeepSeek, server-side Search and Fetch calls are free for now. DeepSeek says billing for those starts on October 31, 2026 at 16:00 UTC.
That makes the next two weeks a cheap time to test it. Run a few of your real queries and compare the results with your current provider before the meter starts.
Who should try it, and a simple way to test it
The best fit is work where volume is high and speed is not critical. Think nightly research jobs, evaluation runs, content checks, or an internal assistant used by a small team. In those cases the low price adds up, and a 5-per-second cap rarely gets in the way.
It is a poor fit for a public app with many people searching at the same moment. It is also a poor fit if your data cannot leave your region. If either applies, a pricier provider is the safer pick.
A fair test takes an afternoon. Here is a quick way to do it:
- Pick 50 real queries your app already sends.
- Run them through DeepSeek and your current provider.
- Compare how many results are actually relevant, and how many retries you needed.
- Check for 400 and 429 errors from long queries or bursts.
Count the cost per useful answer, not per call. A cheap call that returns junk still costs you model tokens to read it.
Frequently Asked Questions
How much does the DeepSeek Search API cost?
It costs $0.75 per 1,000 successful requests, and the Fetch API uses the same rate. Failed or empty calls should not be charged.
Is the DeepSeek Search API free?
Only in one case. Server-side calls through DeepSeek Harness, Claude Code and Codex are free until October 31, 2026 at 16:00 UTC. Everything else is billed.
What does the Fetch API return?
It takes a URL and returns the main text of the page as markdown. That makes it easy to feed into a model for summaries or analysis.
Is it safe to use for private data?
Be careful. DeepSeek’s privacy policy says its servers are in China, so queries and fetched URLs leave your region. Keep sensitive searches on a provider your compliance team has approved.
My take: use it for batch research, testing and low-traffic internal tools, where $60 versus $450 is real money. For anything customer-facing, keep your current provider as the main route and treat DeepSeek as a cheap backup until the quality and the rate limits are proven.
Leave a Reply