DeepSeek drops separate fee for model-triggered web searches and page fetches
From 18 November, DeepSeek no longer charges a separate tool fee for searches and page fetches that the model triggers during a call. Tokens are still billed, and the Search and Fetch APIs that applications call directly still cost 5 yuan per 1,000 successful calls each.
The change took effect at 12:00 Beijing time on 18 November. Pandaily reported it, summarising DeepSeek’s Chinese-language pricing notice, its tool documentation and a report by IT Home.
DeepSeek no longer charges a separate tool fee for server-side tools. With these, the model decides mid-response to search the web for text or images or to fetch a page’s content, and DeepSeek runs the search on its own servers. The change covers DeepSeek’s web tools and the web_search tools used through the Responses API and the Anthropic-compatible API. Input and output tokens are still billed at the model’s prices.
The change does not cover every DeepSeek search. What matters is whether the tool runs inside the model or the application calls a standalone API directly. The standalone APIs /v1/tools/search and /v1/tools/fetch each cost 5 yuan per 1,000 successful calls. An AI agent or other automated application that calls these APIs is not exempt from the fee because the AI decided to make the request, Pandaily said.
The standalone APIs charge for an HTTP 200 response that produces a result: a search must return at least one result, and a fetch must return the page content. DeepSeek does not charge for failed requests, empty searches or timeouts, and the number of results requested does not change the price.
The standalone Search API does not yet support image search. For now, image search works through the model’s web tools, so the price change does not mean a new image search API.
DeepSeek Harness is DeepSeek’s harness, a program that steers an agent by giving the model tools and running its steps. The release notes for the Harness release candidate v0.2.2-rc.5 say DeepSeek models use server-side search with no per-search charge. Other models, however, default to the separately billed DeepSeek Search API. Pandaily reported that Harness uses the model-side route when the search provider is set to server-side search or to Auto, the default. The publication does not say whether this also applies to non-DeepSeek models. In practice, it is worth checking which API or search provider your own integration actually uses, Pandaily said. Harness is available on GitHub.
Read more: How an agent works · Agent map