Baidu Wenxin API — answers and sources from Baidu's China AI chat - querying.ai

BAIDU_WENXIN asks Baidu’s Wenxin chat (文心, wenxin.baidu.com) as a guest. It searches the web before answering, and no account or login is involved.


taskType	`BAIDU_WENXIN`
Reads	`prompt` (falls back to `query`)
Credits	1.25
Country	Ignored. Every task asks from mainland China.

Request

Wenxin answers guests only from mainland China, so every task leaves from a mainland exit and country does not change it. You can omit it.

curl -X POST https://api.querying.ai/v1/async/task \
  -H "Authorization: Bearer $QUERYING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "taskType": "BAIDU_WENXIN", "payload": { "prompt": "上海有哪些适合周末去的博物馆" } }'

Response

Field	Type	Description
`text`	string	The answer.
`markdown`	string	The answer as Markdown. Returned without any flag.
`searchQueries`	string[]	The searches Wenxin ran before answering.
`sources`	array	The pages Wenxin read: `position`, `url`, `label`, `source` (the site name).
`rawResponse`	object	Cards shown with the answer (places, products, images) as `components`. Only with `include.rawResponse`.

The answer text carries no citation marks: sources lists what Wenxin read, not what it cited. A typical answer reads about 30 pages. For example:

{
  "text": "上海适合周末去的博物馆有上海博物馆东馆 ...",
  "markdown": "上海适合周末去的博物馆有**上海博物馆东馆** ...",
  "searchQueries": ["上海 周末 博物馆 推荐"],
  "sources": [{ "position": 1, "url": "https://www.example.com/shanghai-museums", "label": "上海博物馆推荐", "source": "example.com" }]
}

Include flags

Flag	Effect
`rawResponse`	Adds `rawResponse`.

markdown and searchQueries need no flag. ads and shopping have no effect. analysis is rejected with 422 VALIDATION_ERROR, citationAnalysis with 400 VALIDATION_ERROR and html with 422 EVIDENCE_UNSUPPORTED. See Markdown and raw responses.