DeepSeek Search & Fetch API: Help AI Find Information, Read Pages and Check Sources

DeepSeek Search / Fetch integration with QVeris

QVeris has integrated DeepSeek web search and page retrieval. The DeepSeek Platform provider page is the entry point; this review also discovered both tools through QVeris and successfully executed two Search requests and two Fetch requests on October 28, 2026.

For an agent that needs to check a pricing announcement, the workflow is concrete: find candidate sources, inspect the relevant page, and answer with evidence. The integration brings the two retrieval actions into QVeris’s tool-discovery and execution workflow. It does not itself validate every claim in the retrieved text.

This is a live integration check and a small workflow comparison. It is not a benchmark against other search providers or a measurement of DeepSeek’s direct API versus QVeris latency.

What costs ¥5 per 1,000 successful requests?

The official Chinese pricing page lists ¥5 per 1,000 successful requests for Search and the same rate for Fetch. That is ¥0.005 per billable request. Model token charges are a separate cost. All CNY calculations below use this Chinese price schedule; they are not currency conversions from the English price page.

The announced change at 00:00 Beijing time on November 1, 2026 (UTC+8) concerns the previously unbilled Search/Fetch requests made through server-side tool calls when using DeepSeek API in DeepSeek Harness, Claude Code or Codex. As of this review date, that switch is still in the future. It is not a statement that every search performed by those applications uses DeepSeek.

These conditions come from the Search & Fetch guide. In QVeris, use the current tool quote and actual settlement for your platform budget. In our four preflight checks, each request quoted 1 credit; execution responses also recorded 1 credit each but marked settlement pending. This is not a verified final charge or a conversion of ¥0.005.

How the comparison was run

The test question was: “What are DeepSeek Search/Fetch prices and billing conditions?” Four requests ran sequentially through QVeris at approximately 16:17 Beijing time on October 28, 2026. Each configuration was called once. The two Search calls used the identical query and count, changing only need_content.

{
  "query": "DeepSeek API Search Fetch 价格 site:api-docs.deepseek.com",
  "count": 3,
  "need_content": false
}

Run B changed need_content to true. Runs C and D fetched the Chinese pricing page and the Search/Fetch guide respectively. We recorded success, returned URLs, presence of content, character counts and QVeris’s elapsed_time_ms. Character counts include whitespace and Markdown escaping; they are not tokens or a content-quality score.

All four responses had success=true and an upstream status_code of 200. Timing is the field returned by QVeris, not client wall-clock duration or a guarantee of typical latency. The sample is too small to establish performance distributions, availability or the effect of caching.

Case 1: summaries versus page content

Both Search configurations returned the same ordered URLs: the English pricing page, the Chinese pricing page, and a historical context-caching news page. Two of the three were current pricing pages relevant to the task; the historical article was not the primary evidence for current Search/Fetch billing. This is a manual relevance classification of this one result set, not a general accuracy percentage.

The summary-only response already supplied a useful pricing lead. Adding body text made more of each source available, but did not remove the irrelevant historical result. For a brief source-finding step, summaries were sufficient in this test; for checking context, the body text provided additional material to inspect.

A reasonable workflow is to review titles and URLs first, prefer the desired language and canonical source, then inspect bodies only where the question requires detail. Search with body text can reduce separate fetches when it already supplies sufficient evidence, but that benefit depends on actual returned content.

Case 2: a price page versus the full usage guide

We then used Fetch on two known official URLs. Both returned a title and body. We checked four facts relevant to the question: CNY price, the announced server-side billing start time, the result-based billing condition, and whether count changes the price.

The pricing page answered the amount and date; the guide completed the operational conditions. This is why a pricing announcement alone is not enough for an agent that must decide which requests to count in a budget. Known URLs can be fetched directly without first spending a search request.

The returned Fetch text contained escaped Markdown punctuation, so a literal search for an unescaped timestamp could miss it. Normalize formatting for display or keyword checks, while preserving the original response for audit. The larger character count in Fetch does not by itself show that it captured more facts than Search content.

Cost example: count calls, then reduce unnecessary retrieval

The following is a calculation, not a production bill. Assume every request succeeds with billable content, each task is independent, no cached results are reused, and all requests follow DeepSeek’s native CNY tariff. Formula: retrieval cost = (billable Search calls + billable Fetch calls) × ¥0.005.

Under these assumptions, selecting two pages instead of fetching all three reduces retrieval cost from ¥200 to ¥150 per 10,000 tasks, a 25% reduction. This is a strategy estimate, not observed savings and not evidence that QVeris is cheaper than a direct DeepSeek integration. Workflows may also retrieve different evidence, so choose the least costly one that still answers the task.

Token generation, application hosting and QVeris platform settlement are outside this table. More retrieved text can increase downstream model input even when the search request price remains unchanged. Track retrieval calls and model tokens separately.

Use DeepSeek Search and Fetch through QVeris

Open the DeepSeek provider details on QVeris to review the Search and Fetch tool descriptions and usage information. Use Search to find relevant web pages, or Fetch to read the content of a specific page. Choose the capability that fits your task.

Frequently asked questions

Sources and reproducibility

Check the official CNY pricing page, the Search & Fetch guide and the DeepSeek provider entry. Prices and tool contracts were checked on October 28, 2026. Later readers should recheck the current versions before budgeting.

The delivery includes a benchmark.json audit summary with exact inputs, returned URLs, character counts, timestamps and all four execution IDs. It contains no credentials. Replaying requests can produce different results and may consume QVeris credits.

Try a source-backed billing check

Use the task below to repeat the workflow with current data. The button opens a prefilled Playground prompt for you to review and send.