Model Studio Qwen Deep Research

Validation

mkdir -p output/aliyun-qwen-deep-research
python -m py_compile skills/ai/research/aliyun-qwen-deep-research/scripts/prepare_deep_research_request.py && echo "py_compile_ok" > output/aliyun-qwen-deep-research/validate.txt

output/aliyun-qwen-deep-research/validate.txt

Output And Evidence

• Save research goals, confirmation answers, normalized request payloads, and final report snapshots under

  output/aliyun-qwen-deep-research/

  .
• Keep the exact model, region, and

  enable_feedback

  setting with each saved run.

Critical model names

• qwen-deep-research

• qwen-deep-research-2025-12-15

• Use

  qwen-deep-research

  for the current mainline model.
• Use

  qwen-deep-research-2025-12-15

  when you need the snapshot with MCP tool-calling support and stronger reproducibility.

Prerequisites

• Install SDK in a virtual environment:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install dashscope

• Set

  DASHSCOPE_API_KEY

  in your environment, or add

  dashscope_api_key

  to

  ~/.alibabacloud/credentials

  .
• This model currently applies to the China mainland (Beijing) region and uses its own API shape rather than OpenAI-compatible mode.

Normalized interface (research.run)

Request

• topic

  (string, required)

• model

  (string, optional): default

  qwen-deep-research

• messages

  (array<object>, optional)

• enable_feedback

  (bool, optional): default

  true

• stream

  (bool, optional): must be

  true

• attachments

  (array<object>, optional): image URLs and related context

Response

• status

  (string): stage status such as

  thinking

  ,

  researching

  , or

  finished

• text

  (string, optional): streamed content chunk

• report

  (string, optional): final structured research report

• raw

  (object, optional)

Quick start

python skills/ai/research/aliyun-qwen-deep-research/scripts/prepare_deep_research_request.py \
  --topic "Compare cloud video generation model trade-offs for marketing automation." \
  --disable-feedback

Operational guidance

• Expect streaming output only.
• Keep the initial topic concrete and bounded; broad topics can trigger long iterative search plans.
• If the model asks follow-up questions and you already know the constraints, answer them explicitly to avoid wasted rounds.
• Use the snapshot model when you need stable evaluation runs or MCP tool-calling support.

Output location

• Default output:

  output/aliyun-qwen-deep-research/requests/

• Override base dir with

  OUTPUT_DIR

  .

References

• references/sources.md