DeepSeek's API has built-in web search — freeload the official search via the Responses API

Originally posted on X. This one blew up — 230k views, 1000+ likes. I'll expand on the details here.

Found something great: DeepSeek actually shipped web search directly inside its API.

In one sentence

Call the deepseek-v4-flash model through the Responses interface. Just declare a web_search tool in the request parameters, and you get DeepSeek's server-side search capability.

No integrating a third-party search engine, no applying for a separate search API key — the entire search pipeline is hosted by DeepSeek.

How to use it

Official docs: https://api-docs.deepseek.com/zh-cn/guides/responses_api

The core is just declaring the tool:

const response = await fetch('https://api.deepseek.com/responses', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'Authorization': `Bearer ${process.env.DEEPSEEK_API_KEY}`
  },
  body: JSON.stringify({
    model: 'deepseek-v4-flash',
    tools: [{ type: 'web_search' }],
    input: 'What major things happened in the AI industry in August 2026?'
  })
})

That's it. The searching, fetching, parsing, and citing are all done on DeepSeek's side.

Why this matters

Previously, to give an AI real-time information, you had to build the whole pipeline yourself: pick a search engine (SerpAPI, Bing Search, whatever) → apply for an API key → write the fetch-and-parse code → stuff the results into context → keep an eye on the budget.

That's a day or two at minimum, maybe a week if you hit snags, and every step costs money: search APIs bill per request, and scraping means fighting anti-bot measures.

Now DeepSeek just built it in, using the absurdly cheap deepseek-v4-flash model. Search and generation in one pipeline, cheap enough to use like running water.

Good use cases

• Writing anything time-sensitive (industry news, product comparisons, policy explainers)
• Building agents: steps that need to look things up before deciding
• Customer support / Q&A systems: search the latest info before answering
• Any scenario where the model's knowledge cutoff holds you back

Notes

1. Use the Responses interface, not the old Chat Completions interface. The old one doesn't have this tool.
2. The granularity of web search and the citation format are in the official docs — worth running it once to see the response structure.
3. There's a cache discount mechanism — make use of it for long-context scenarios, it saves a lot.

This is a genuine freebie: the official team gave away the most painful part of search infrastructure. If you need it, just copy the homework.

Related: DeepSeek V4 Flash — hands-on review of the stable release