# DeepSeek Search and Fetch Pricing: What Changes After the Free Period

DeepSeek's free search and fetch calls inside coding agents reach their billing deadline, adding a cost that token prices alone do not show.

By Paras Madan | 2026-11-01


DeepSeek's published free period for Search and Fetch calls through DeepSeek Harness, Claude Code and Codex ended at midnight Beijing time on November 1. Its stated rate is now $0.75 per 1,000 successful requests for each service, separate from model token charges.

That deadline was October 31 at 16:00 UTC, or 9:30 p.m. in India. This is an existing pricing transition taking effect, not a new model launch. The practical change is that an agent's browsing now needs its own line in the cost estimate.

## DeepSeek Search and Fetch prices

The [official pricing page](https://api-docs.deepseek.com/quick_start/pricing/) distinguishes model usage, charged by tokens, from these standalone tools, charged by successful requests.

| Service | Published price | What it returns |
| --- | --- | --- |
| Search | $0.75 per 1,000 successful requests | Ranked web results |
| Fetch | $0.75 per 1,000 successful requests | A page's content |

**The free-period deadline concerns server-side calls through the named agent integrations.** It should not be read as a claim that every standalone API request was previously free. DeepSeek already listed a per-request price for the standalone endpoints.

The documentation still contains wording that those integrated calls are currently free, immediately followed by the explicit November 1 start time. The deadline has passed; the rate here follows that published schedule, not a test of an individual account's invoice.

## What counts as a billable request?

DeepSeek's [Search and Fetch guide](https://api-docs.deepseek.com/guides/search/) requires both an HTTP 200 response and usable output. Search must return a non-empty results array, while Fetch must return page content. Failed requests, empty search results and timeouts are not billed.

Search can return between one and 50 results per request. Changing that count does not change the request price. Search and Fetch have independent account-level limits of 100 and 25 requests per second, with an additional per-domain limit for Fetch.

The distinction matters when estimating research workflows. Finding ten pages is not the same as separately fetching all ten. Count the actual operations your agent performs rather than treating every user question as one search.

## How the APIs fit into an agent

The [Search API](https://api-docs.deepseek.com/api/search/) returns structured results, including URLs, summaries and optional page content. Its `need_content` field defaults to false. A developer can request content with search results, although the response may still contain an empty content field when that content is unavailable.

The [Fetch API](https://api-docs.deepseek.com/api/fetch/) takes a page URL and returns its title and Markdown content. That is useful when an agent needs to read a particular source rather than discover candidates.

Requesting search content could avoid some separate fetches, but that is a workflow choice to measure, not a guarantee that every retrieved page will be complete.

## What this means for agent costs

Consider an illustrative workflow with two successful searches and eight successful fetches. Its ten tool requests cost $0.0075. Running that same pattern 100,000 times costs $750 in these tool fees, before model token charges or other infrastructure.

**The useful metric is cost per completed task.** A cheaper search loop that repeatedly retrieves the wrong pages can be worse than a slightly longer one that reaches a reliable answer. Track successful calls, repeated URLs and task completion together. The billing change makes that distinction visible instead of leaving browsing inside an apparently free part of the agent.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders turn product launches and engineering updates into clear stories developers can understand and share.

[Book a 30-minute call](https://cal.com/paras-madan/30min).