Alibaba Cloud AI Code Qwen Coder

Category: provider

Model Studio Qwen Coder

Validation

mkdir -p output/aliyun-qwen-coder
python -m py_compile skills/ai/code/aliyun-qwen-coder/scripts/prepare_code_request.py && echo "py_compile_ok" > output/aliyun-qwen-coder/validate.txt

Pass criteria: command exits 0 and output/aliyun-qwen-coder/validate.txt is generated.

Output And Evidence

• Save prompts, repository context summaries, and normalized coding request payloads under output/aliyun-qwen-coder/.
• Record the exact model, endpoint mode, and target language/framework for reproducibility.

Use Qwen Coder for coding assistants, code review drafting, repository-level reasoning, patch planning, and tool-using software agents.

Critical model names

Use one of these exact model strings as appropriate:

• qwen3-coder-next
• qwen3-coder-plus
• qwen-coder-plus
• qwen2.5-coder-32b-instruct

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.
• Prefer the OpenAI-compatible endpoint when a client does not yet expose the latest coder models in dropdown UI.

Normalized interface (code.generate)

Request

• messages (array, required)
• model (string, optional): default qwen3-coder-next
• repository_summary (string, optional)
• files (array, optional)
• language (string, optional)
• tools (array, optional)
• stream (bool, optional)

Response

• text (string)
• patch (string, optional)
• usage (object, optional)

Quick start

python skills/ai/code/aliyun-qwen-coder/scripts/prepare_code_request.py \
  --task "Refactor request validation into a small helper and add one unit test." \
  --language python

Operational guidance

• Pass only the files relevant to the requested change to reduce noise.
• Use qwen3-coder-next for current-generation coding tasks and qwen3-coder-plus when you need a stronger but potentially costlier coder.
• For repo-scale changes, include architecture notes and expected test commands.
• Prefer deterministic prompts and pinned model IDs for benchmarking or regression comparison.

Output location

• Default output: output/aliyun-qwen-coder/requests/
• Override base dir with OUTPUT_DIR.

References

• references/sources.md