DeepSeek (Search & Page Fetch, China Region Only)
The platform offers two DeepSeek endpoints: Search for web search results, and Fetch for turning a single web page into Markdown. They are independent — call them on their own, or chain them (Search to find pages → Fetch for content → feed to an LLM).
Official documentation: DeepSeek Search & Fetch ↗
Common conventions
- Available in the China region only; calls from any other region return
404(error code 1003). - Both endpoints are
POST+ JSON body and need only one header:Authorization: Bearer $TURING_API_KEY. - Each endpoint accepts only the fields listed in the tables below. Any other field returns
422and the request is never sent. - Fields whose value is
nullare omitted from the response.
Search
- Endpoint:
POST /proxy/deepseek/search - Required parameter:
query
| Parameter | Type / default | Description |
|---|---|---|
query | string, required | Query string, 1–128 characters, not all whitespace. Supports the search syntax below |
count | int, default 10 | Number of results, 1–50. Changes only how many results come back, not the price |
ori_question | string | The user's original question, 1–128 characters, to help interpret the search intent |
localization | string | An IP address (IPv4 / IPv6); results local to that area are ranked first |
need_content | bool, default false | Whether each result carries the page body in content |
Search syntax supported in query; operators can be combined in one query:
| Syntax | Effect |
|---|---|
site:example.com | Search only the given site; join several sites with OR |
-site:example.com | Exclude the given site |
"exact phrase" | Match the phrase exactly |
-keyword / -"phrase" | Exclude results containing the keyword or phrase |
lang:zh,en | Restrict result languages, comma-separated |
after:2026-01-01 / before:2026 | Restrict publication date, as a year or YYYY-MM-DD |
curl $TURING_BASE_URL/proxy/deepseek/search \
-H "Authorization: Bearer $TURING_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "DeepSeek API site:api-docs.deepseek.com",
"count": 3
}'
Response excerpt:
{
"results": [
{
"title": "Change Log | DeepSeek API Docs",
"url": "https://api-docs.deepseek.com/updates/",
"snippet": "The official V4-Flash natively supports the Responses API format…",
"summary": "## Date:\n...\nFor more details, please refer to this documentation…",
"content": "",
"site_name": "DeepSeek",
"site_icon": "https://cdn.deepseek.com/site-icons/deepseek.com",
"date": "2026-09-09T16:00:00+00:00",
"authority": 3,
"rerank_score": 2.26
}
]
}
The top level holds a single results array, which is empty when nothing matches. Each result has these fields:
| Field | Description |
|---|---|
title / url | Page title and address |
snippet | Short summary, usually 100–200 characters, suited to display in a UI |
summary | Long summary, usually 600–800 characters, suited to feeding a model directly |
content | Page body; an empty string when need_content is false or the page has no body |
site_name / site_icon | Site name and icon URL; omitted when unknown |
date | Publication time (UTC, RFC 3339); omitted when unknown |
authority | Site authority, 0–4, with 3 and above being highly authoritative; -1 means unknown |
rerank_score | Relevance to the query, 0–3; 1.5 and above can be treated as high quality |
Fetch
Fetches one web page and converts it to Markdown — useful for reading the full body of a URL found with Search.
- Endpoint:
POST /proxy/deepseek/fetch - Required parameter:
url
| Parameter | Type / default | Description |
|---|---|---|
url | string, required | The page to fetch, ≤8192 characters; must be an http(s) address with a host name, and must not carry a username or password |
curl $TURING_BASE_URL/proxy/deepseek/fetch \
-H "Authorization: Bearer $TURING_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"url": "https://api-docs.deepseek.com/"
}'
Response excerpt:
{
"title": "Your First API Call | DeepSeek API Docs",
"content": "…# Your First API Call\n\nThe DeepSeek API uses an API format compatible with OpenAI/Anthropic…"
}
| Field | Description |
|---|---|
title | Page title; an empty string when there is none |
content | Page body in Markdown |
Whether a page can be fetched depends on the target site's availability and access policies. When the page is unreachable, blocked by the site, of an unsupported content type, or yields no extractable body, the request fails with 502.
Billing
Billed per request; Search and Fetch cost the same, ¥0.005 per request (¥5 per 1,000). Only successful requests that produce a result are billed:
| Endpoint | Billing rule |
|---|---|
| Search | 1 request when at least 1 result comes back, regardless of count or the actual number of results; an empty array is not billed |
| Fetch | 1 request when page content comes back; empty content is not billed |
Failed requests — validation errors, upstream failures, invalid responses — are not billed.
- Rate limits: Search / Fetch are counted separately, at a default of 480 requests per hour per endpoint (enterprise customers are provisioned separately). Quota and rate limits are checked before the request goes out.
404(error code 1003): Called from a region other than China.422: The request body failed a field or value check; nothing was sent.429(error code 3091): Upstream rate limit hit; wait and retry. Fetching the same site many times in a short period can also trigger it.502(error code 3090): Upstream error, Fetch could not get usable content, malformed response, or a response body larger than 10 MB. A uniform error message is returned rather than the upstream's raw error content, so include theX-Turing-Trace-Idwhen reporting an issue.
See Error Codes for error code meanings and Rate Limits for the rate-limiting policy.