DeepSeek API Charges 5 Yuan per 1000 Web Searches
DeepSeek announced that its API now supports web search and scraping functions, charging based on the number of successful requests, with a price of 5 yuan for every 1000 successful requests.
Currently, users calling the DeepSeek API through DeepSeek Harness, Claude Code, or Codex have accessed the Search and Fetch interfaces via server-side tools, and this part of the call is currently free of charge. Starting from 00:00 Beijing time on November 1, 2026, it will begin charging per request.
The Search interface is powered by DeepSeek's self-developed search engine, returning structured web results based on queries; the Fetch interface retrieves a single webpage and returns the title and content in Markdown format. Both are independent JSON APIs that use the same authentication method as model calls.
Charges will only apply when the request returns HTTP 200 and produces valid results (Search requires a non-empty result list, Fetch requires returning page content). Failed requests, empty results, and timeouts will not incur charges, and the number of returned results does not affect the price per request.
Each account is limited to 5 requests per second for both Search and Fetch, calculated independently and not affecting the concurrent limits of model calls. Fees will be directly deducted from the account balance.
In the market, developers and agent application providers will bear additional retrieval costs, with funds flowing from users to DeepSeek. High-frequency web tool calling scenarios will be under pressure, while providers relying on their own search capabilities will benefit, and the cost structure for pure model inference remains unchanged.
Supplementary information indicates that this feature was officially launched recently, providing real-time web content for agents and RAG applications. Currently, server-side calls through designated coding tools are still in a free trial period.
Source: Public Information
ABAB AI Insight
DeepSeek previously attracted developer and application traffic quickly through low-cost large model APIs and provided web search capabilities within applications. This time, it opens its self-developed search engine capabilities as an independent API, continuing its path of expanding from model inference to tool layers, while introducing separate charges for high-frequency tool calls.
The capital path reflects first attracting agent and coding tool users to use the server-side Search/Fetch through free or low-cost services, then converting successful requests into per-request revenue after a fixed date, with funds flowing directly from developer accounts to the platform, prioritizing calls with actual results over failed requests.
Similar cases can be seen in the practices of OpenAI, Google, and others gradually charging for search or tool calls, as well as early search API providers charging per request or result. Current AI applications are in a stage of expanding from pure model calls to tool integration and real-time retrieval, with retrieval costs shifting from external services to monetization of the platform's own capabilities.
Essentially, this represents a transfer of pricing power: the platform separates previously bundled or free retrieval capabilities into independent chargeable items, making retrieval costs explicit through charging for successful requests, prompting developers to optimize call frequency and result utilization, while reinforcing the entry position of its own search engine in agent workflows.
ABAB News · Cognitive Laws
First cultivate habits for free, then charge per request.
The more tool calls, the easier it is to shift pricing power.
Only successful requests incur charges, while the platform bears the cost of failures.