Back to the feedDeepSeek Credits

Was $0.75 per 1,000 requests

DeepSeek API now runs model web search and page fetch for free, you only pay for tokens

Model-triggered web search and page fetch are no longer billed per call; you only pay for tokens

Since 12:00 Beijing time on November 18, 2026, server-side web search and page fetch calls the model makes on the DeepSeek API are no longer billed per request. They used to cost $0.75 per 1,000 successful requests. Results are still billed as model tokens, and Claude Code's Web Search and DeepSeek Harness's default search are covered.

Eligible plans
DeepSeek API pay-as-you-go
Eligibility
All DeepSeek API users; calling the Search or Fetch API directly is still billed per request

The briefing

DeepSeek has changed how web search is billed. Starting 12:00 on November 18, 2026 (Beijing time, UTC+8), the server-side web search and page fetch calls the model makes on its own are no longer billed per call. They used to cost $0.75 per 1,000 successful requests; now you only pay for the model tokens you use.

What you get

  • Text search, image search and page fetch calls that the model runs through server-side tools are no longer charged per request.
  • This covers the DeepSeek API web tool, the web_search tool in the Anthropic-format API and the Responses API, Claude Code's Web Search, and DeepSeek Harness searches when "Web search provider" is set to "DeepSeek server-side search" or "Auto" (the default).
  • The search results are added to the model context and are still billed as model token usage.
  • Calling the Search or Fetch API directly from your own code is still billed at $0.75 per 1,000 successful requests. That also applies when DeepSeek Harness uses "DeepSeek Search API" as its search provider, and to Codex Alpha Search.

How to use it

  1. Sign in to the DeepSeek Platform, create an API key, and make sure your account has a balance.
  2. Point Claude Code or DeepSeek Harness at the DeepSeek API. In DeepSeek Harness, leave "Web search provider" on the default "Auto".
  3. Ask questions as usual. When the model needs the web, it searches and fetches pages on its own, and you only pay for its tokens.

Billing details are on DeepSeek Models & Pricing and in the Search & Fetch guide.

Is this still live?

Community signal100% say it still works