changelogDeepSeek

DeepSeek removes separate fees for model-triggered search and web fetch from Nov 18

TL;DR

DeepSeek removed separate charges for server-side search and web fetch triggered during model calls, effective at noon on November 18, 2026, Beijing time. Token usage is still billed, and the standalone Search and Fetch APIs remain at 5 yuan per 1,000 successful calls.

2 min read
0

DeepSeek has removed separate charges for server-side search and web fetch triggered during model calls, effective at noon on November 18, 2026, Beijing time (UTC+8). Its Chinese pricing documentation says users still pay for model-token usage. The change does not make all DeepSeek search services free.

What changed

The update separates tools that run inside a model interaction from standalone requests made directly by an application's client. DeepSeek's Chinese search-and-fetch guide lists text search, image search and page fetch as model-triggered server tools that carry no separate tool fee.

The pricing notice covers DeepSeek's web tools and the web_search tools exposed through its Responses API and Anthropic-compatible API. On those model-side routes, the tool charge is gone, but input and output tokens remain billable under the applicable model pricing.

What still costs money

Direct client calls to the standalone Search and Fetch APIs keep their per-request prices:

  • Each API costs 5 yuan per 1,000 successful calls.
  • The endpoints are /v1/tools/search and /v1/tools/fetch.
  • An agent or automated application calling those endpoints is still charged, even if an AI decided to make the request.

DeepSeek defines a billable standalone request as an HTTP 200 response that produces a result. Search must return a nonempty results array, and Fetch must return page content. Failed requests, empty searches and timeouts are not charged. According to the official guide, the number of search results requested does not change the per-request price.

Configuration example

The documentation uses DeepSeek Harness as an example. When its web-search provider is set to DeepSeek server-side search or Auto (the default), it uses the model-side route. Choosing DeepSeek Search API instead uses the separately billed client route.

Image search limitation

The standalone Search API does not yet support image search. The guide says image search is currently available through the web tools. The pricing change is therefore not a new standalone image-search endpoint.

Context

This is a billing-policy update to existing tool paths. It is distinct from the October 17 launch of independent Search and Fetch APIs and from the broader application-API announcement covered in November. The reporting draws on DeepSeek's Chinese pricing notice and Chinese billing and tool guide, as well as an IT Home report on the pricing update, via Pandaily.

What this means

The bill for a search-enabled workflow now depends on which route it takes. A call that goes through the Responses API, the Anthropic-compatible API or DeepSeek's web tools incurs token costs only. The same search made through /v1/tools/search or /v1/tools/fetch still costs 5 yuan per 1,000 successful calls.

For developers, the practical check is which endpoint or provider the integration actually uses, not whether the workflow is described as an AI search agent. In DeepSeek Harness, the provider setting decides which billing path applies, and the default (Auto) uses the model-side route. Teams that need image search should note that it is only available through the web tools for now.

Comments

Loading...