,

DeepSeek Web Search API: $0.75 per 1,000, 13x Cheaper

By

·

·

7–11 minutes

The DeepSeek web search API now costs $0.75 per 1,000 searches, according to DeepSeek’s pricing page, after the company opened its app’s tools to developers on November 13. OpenAI and Anthropic charge $10 per 1,000 for their built-in search tools, and Google charges $14 past a free tier. That makes DeepSeek about 13 times cheaper per search, with file parsing and voice priced just as low.

Key takeaways

  • DeepSeek charges $0.75 per 1,000 web searches, 92.5% below the $10 OpenAI and Anthropic charge.
  • One million searches a month costs $750 on DeepSeek versus $13,930 on Gemini grounding.
  • The tools run only on deepseek-flash, priced at $0.30 input and $1.20 output per million tokens.

What did DeepSeek launch?

DeepSeek turned the machinery behind its consumer chat app into paid API features. Developers can now switch on web search, deep-thinking search, file input, URL reading and safety checks inside the existing API, or call search, fetch, OCR, speech recognition and text-to-speech as standalone services. Every tool is billed per use, on top of tokens.

The announcement is dated November 13, 2026. DeepSeek says the tool calls and context assembly run on its own servers, so a developer flips a parameter instead of wiring up a search provider.

The pitch is blunt. DeepSeek tells builders to “spend less time rebuilding the basics and more time creating something new.” It also published an open-source chat demo that can import a user’s DeepSeek conversation history.

Which tools are included?

  • Web search: text or image search with citations, in non-thinking mode.
  • Deep thinking search: the model searches and reads pages across multi-step reasoning.
  • File input: PDFs, Office files and images parsed into context.
  • URL reading: links in a user’s message are fetched automatically.
  • Safety checks: inputs and outputs are screened, with refusal flags returned when risk is detected.

All of it works across three API formats: Messages, Responses and Chat Completions. That matters for switching costs. A team already calling OpenAI-style or Anthropic-style endpoints can point the same code at DeepSeek.

How much does the DeepSeek web search API cost?

DeepSeek lists search, for text or images, at $0.75 per 1,000 searches and web fetch at $0.75 per 1,000 fetches. The fee is the same whether a developer calls the standalone endpoint or the model triggers the search itself. Token charges for the pages the model reads come on top, at deepseek-flash rates.

Those token rates are low too. DeepSeek’s pricing page puts deepseek-flash at $0.30 per million input tokens and $1.20 per million output tokens at peak, and half that off-peak.

Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays. Every hour after 10:00 UTC, which is 5 a.m. in New York, is off-peak. In practice, a US company running agents during its own business day pays the discounted $0.15 input and $0.60 output rate.

DeepSeek tool price list

ToolUnitPrice
Search (text, image)1,000 searches$0.75
Web fetch1,000 fetches$0.75
File parsing / OCR1,000 pages$0.225
Text to speech10,000 characters$0.45 list, $0.225 limited-time
Speech to text1 hour of audio$0.116 list, $0.058 limited-time
deepseek-flash tokens (peak)1M input / 1M output$0.30 / $1.20
Source: DeepSeek API pricing page, checked November 15, 2026.

Is DeepSeek cheaper than OpenAI, Claude and Gemini for web search?

Yes, by a wide margin on the per-search fee. Anthropic’s pricing docs list Claude’s web search at $10 per 1,000 searches plus tokens. OpenAI’s current web search tool also costs $10 per 1,000 calls on every model, according to a November 5 breakdown of OpenAI’s developer pricing by search provider Parallel.

Google’s Gemini API pricing includes 5,000 grounded search queries a month for free on Gemini 3 models, then charges $14 per 1,000. Even the cheapest independent search API we checked, Parallel’s Turbo and Fast modes, sits at $1 per 1,000 requests.

ProviderPer 1,000 searchesCost of 1M searches a monthDeepSeek’s discount
DeepSeek (built-in or standalone)$0.75$750—
Parallel Search (Turbo/Fast)$1.00$1,00025%
OpenAI web_search tool$10.00$10,00092.5%
Claude web search tool$10.00$10,00092.5%
Gemini grounding (after 5,000 free)$14.00$13,93094.6%
Search fees only; token charges are extra on every platform. Monthly totals are Wealth Engine’s arithmetic from the listed prices.

What does one agent task cost?

Take a simple research step: three searches, 20,000 tokens of pages read and a 1,000-token answer, run during US hours. On DeepSeek, the searches cost $0.00225, the input $0.003 and the output $0.0006. The whole task lands near six-tenths of a cent.

The same three searches through OpenAI’s or Claude’s tool cost $0.03 in search fees alone, before a single token. For agents that search thousands of times an hour, that gap decides whether the product has a margin.

Search fees are not a rounding error elsewhere either. When Artificial Analysis benchmarked OpenAI’s web search tool, it measured $40.19 in search fees per 1,000 tasks against $9.40 in model tokens, Parallel reported. On that test, search was the bigger line.

What is the catch?

Price is the only thing DeepSeek has proven so far. We found no independent benchmark of its search quality, and a cheap search that returns stale or thin results can cost more in extra calls and wrong answers than a pricier one. Teams should test before they migrate.

The prices are also not locked. DeepSeek’s own page says product prices may vary and that it reserves the right to adjust them. That warning has history: Wealth Engine covered a DeepSeek price increase of up to 1,100% in August. A developer who rebuilds an agent stack around $0.75 searches is betting the number holds.

There are operational limits, too:

  1. One model only. The built-in tools run on deepseek-flash. The pricier deepseek-v4-pro is not part of the launch.
  2. Mandatory safety checks. Public-facing apps must enable DeepSeek’s screening, which returns refusals when it flags risk. That can block answers a US product would allow.
  3. Compliance stays with you. DeepSeek says developers remain responsible for content moderation, model registration where it applies and labeling AI-generated content.
  4. Data jurisdiction. Every search query and fetched page passes through a China-based provider. For regulated US firms, that alone may rule it out.

Who wins and who loses?

Cost-sensitive agent builders win first. Independent search API vendors lose the most, because DeepSeek undercuts even the $1 floor they have marketed as the cheap alternative. The big labs lose pricing power on a tool fee that had quietly become one of the largest lines in an agent bill.

Winners

Startups running high-volume agents. Customer support bots, price trackers and research agents search constantly. A 92.5% cut on the search line can move a product from loss to profit.

DeepSeek itself. It already ran this infrastructure for its consumer app. Selling it by the call turns a cost center into revenue and pulls developers deeper into its API, after the V4.1 Flash 1017 update already cut per-task costs.

Losers

Search API startups. Parallel, Tavily, Brave and Exa sell search to agents as a standalone product. DeepSeek bundles comparable functions at a lower price, with the model included.

OpenAI, Anthropic and Google. Their $10 to $14 tool fees now look like margin, not cost. Google has the most to protect, since grounding fees sit next to its search advertising business. We saw the same pressure on model prices this week with Gemini 4 Hydrogen’s pricing.

Our view: expect the US labs to defend on quality and compliance, not to match $0.75. Their enterprise buyers pay for data residency and legal comfort, which DeepSeek cannot offer them.

What to watch next

DeepSeek published no dates for the next steps, so the signals matter more than a calendar. Four are worth tracking over the coming weeks.

  • The end of the speech discount. TTS and speech recognition are 50% off for a limited time, with no end date given. The list price is the real one to budget.
  • A rival fee cut. Any move by OpenAI, Anthropic or Google on their $10 to $14 search fees would confirm DeepSeek reset the market.
  • Independent quality tests. A benchmark such as Artificial Analysis’s search index adding DeepSeek would show whether the cheap searches are good searches.
  • Tool support on V4 Pro. Extending the tools beyond deepseek-flash would make the bundle relevant for harder tasks.

FAQ

How much does a DeepSeek web search cost?

$0.75 per 1,000 searches, or $0.00075 each, plus token charges for the content the model reads.

Which DeepSeek model supports the built-in tools?

Only deepseek-flash, which runs DeepSeek-V4.1-Flash-1017. The standalone search, fetch, OCR and speech APIs can be used with any workflow.

Is DeepSeek web search cheaper than OpenAI’s?

Yes. OpenAI’s web search tool costs $10 per 1,000 calls, so DeepSeek is 92.5% cheaper on the search fee.

When are DeepSeek’s off-peak hours?

Everything outside 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, plus weekends and Chinese public holidays. Token prices are half during off-peak.

Does DeepSeek charge more when the model triggers the search?

No. DeepSeek bills search and fetch per use at the same rate whether the model or the developer triggers them.

Is the DeepSeek chat demo open source?

Yes. DeepSeek published the frontend demo on GitHub, and it can import a user’s DeepSeek conversation history.

The bottom line

DeepSeek has made web search the cheapest part of an AI agent. At $0.75 per 1,000 searches, it undercuts OpenAI and Anthropic by 92.5% and beats even the discount search vendors.

For prototypes and high-volume consumer agents, it is worth testing now. For regulated companies, the price does not offset the jurisdiction risk. The bigger effect is on everyone else’s pricing: a $10 search fee is now hard to defend.

Free guide AI Tools Every Investor Should Use in 2026 shown as a black and gold book and tablet

Free guide · 12 pages

AI Tools Every Investor Should Use in 2026

10 tools for research, stock signals, charts and crypto, what each one costs, three ready-made stacks from $0 a month and 7 copy-paste prompts for filings and earnings calls. Enter your email and we will send it to you.

Plus one email a week: AI Money This Week, every Sunday. Unsubscribe anytime.

Sources

Wealth Engine researches and drafts with AI tools and checks every figure against the sources above. How we report.

Comments

Leave a Reply