Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableX: @deepseek_ai→ original

DeepSeek Opens Web Search and Page Fetch to Everyone for Agents at $0.75 per 1,000 Calls

DeepSeek has opened two web tools for AI agents to all developers: Search queries the internet, and Fetch pulls the text of a specific page. For example, a bot can look up the changelog of a new library version, read it, and answer based on fresh data rather than memory. Each endpoint costs $0.75 per 1,000 calls, or $0.00075 per call. Teams running agents should consider whether this could replace third-party search.

AI-processed from X: @deepseek_ai; edited by Hamidun News
DeepSeek Opens Web Search and Page Fetch to Everyone for Agents at $0.75 per 1,000 Calls
Source: X: @deepseek_ai. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceX: @deepseek_ai ↗
Editor-in-Chief
Key metric #1Search and Fetch are billed separately, at $0.75 per 1,000 calls each
Key metric #2DeepSeek says its own search engine handles hundreds of millions of requests per day
Key metric #3Calls run as server-side tools, meaning DeepSeek's platform performs the search

DeepSeek has opened two web tools for AI agents to all developers through its API. Search looks up information on the internet, while Fetch retrieves the content of a specific page by its address. Suppose an assistant needs to explain what changed in a new library version: Search finds the changelog page, Fetch loads its text, and the model answers based on fresh data rather than what it memorized during training. Both tools cost the same: $0.75 per 1,000 calls, or $0.00075 per call.

What was launched

Anyone who has built an agent with internet access knows that the model is only half the job. The other half is where it gets fresh data. Typically, a developer connects a third-party search provider, sets up page loading separately, tracks two bills, and watches rate limits. DeepSeek offers to handle both steps from a single provider.

As the company announced in its X account, both APIs are now available to anyone. They run on DeepSeek's own search engine. According to the company, this engine serves hundreds of millions of requests per day worldwide. The announcement contains no independent measurements of result quality, so the load figure should be read as a claim about infrastructure maturity, not as proof of better search quality.

The announcement says the Search and Fetch calls are available as server-side tools in DeepSeek. Judging by the wording, search and page loading run on the platform side, so developers do not need to write their own scaffolding for each step. The response formats, limits, and restrictions are described in the documentation that the post links to.

How much it costs

The pricing is simple: $0.75 per 1,000 calls for each endpoint. That is $0.00075 per call, and a million calls would cost $750. If an agent performs one search and opens two pages for a task, that is three calls, or about $0.00225 per task. A thousand such tasks cost roughly $2.25 for web access alone, excluding the token charges for the model itself.

The original post does not give prices from other providers, so each team will have to decide for itself whether this is cheaper or more expensive by comparing the rates of the services it uses now. Still, a per-call price makes budgeting straightforward: costs grow linearly with the number of calls, and the announcement mentions no hidden surcharges.

What teams should do

  • If you build agents on DeepSeek models, try replacing your third-party search service and compare answer quality and total cost per task on your real use cases.
  • Factor in both endpoints: Fetch is billed separately, and an agent that actively reads pages will spend as much on it as on search.
  • Cap the number of calls per task. An agent without a limit can get stuck in a search loop and run up the bill.
  • Review the legal and privacy aspects: user queries will be sent to a Chinese company's service, which matters for some industries.

Why this matters for the market

AI labs increasingly sell not just models but a set of tools around them. Search is the most in-demand of these tools, because it turns a chatbot into a researcher that can work with current information. When a model provider offers its own search at a fixed price, independent search APIs for agents gain another serious competitor, and developers get fewer intermediaries between the model and the data.

The main question now is not price but quality: how complete and current DeepSeek's search results are, how it handles non-English queries, and how stable it stays under load. The first independent tests by developers will answer that.

Story

Key Questions Answered

What was launched?

Anyone who has built an agent with internet access knows that the model is only half the job. The other half is where it gets fresh data. Typically, a developer connects a third-party search provider, sets up page loading separately, tracks two bills, and watches rate limits. DeepSeek offers to handle both steps from a single provider. As the company announced in its X account, both APIs are now available to anyone. They run on DeepSeek's own search engine.

How much it costs?

The pricing is simple: $0.75 per 1,000 calls for each endpoint. That is $0.00075 per call, and a million calls would cost $750. If an agent performs one search and opens two pages for a task, that is three calls, or about $0.00225 per task. A thousand such tasks cost roughly $2.25 for web access alone, excluding the token charges for the model itself.

What teams should do?

If you build agents on DeepSeek models, try replacing your third-party search service and compare answer quality and total cost per task on your real use cases. Factor in both endpoints: Fetch is billed separately, and an agent that actively reads pages will spend as much on it as on search. Cap the number of calls per task. An agent without a limit can get stuck in a search loop and run up the bill.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…