Dashscope LLM Loader

Don't let "Loader" fool you. The Dashscope LLM Loader holds no weights, uses no VRAM, and makes no API call. It's a dropdown that hands you a Qwen model ID as a string so you don't have to type qwen-max-2024-09-19 from memory in the middle of a workflow. That's its entire existence, and it's exactly what you want from it.

It exists to feed DashscopeModelCaller, the node that actually sends your prompts to Alibaba's DashScope API. That node's model_version input is a plain string with no default - leave it empty and nothing works. Wire this loader's output into it and you get a friendly picklist covering the whole Qwen lineup:

• qwen-max (the default) - the flagship. Strong general reasoning and writing, ~32k context.
• qwen-plus - the value pick, ~131k context.
• qwen-turbo - fast and cheap, with a 1M context.
• qwen-long - the absurd one, 10M context, for when your "prompt" is half a book.

Those context numbers come straight from the pack's own model docs. The practical takeaway: for a normal prompt-rewrite or caption job, qwen-plus or qwen-max is plenty; qwen-turbo shines if you're looping it many times; qwen-long is overkill unless you genuinely need to chew through huge text.

How it works

Same trick as its vision sibling. The node reads model_versions/llm.txt, exposes the entries as an enum, and returns your selection as a STRING. No compute, no cost, no key required - the loader itself is free, and it doesn't even check that DASHSCOPE_API_KEY is set.

Inputs and output

• model_version - the only input, 24 choices across the qwen-max, qwen-plus, qwen-turbo and qwen-long families. Default is qwen-max.

Output is a single STRING carrying the model ID. Wire it into DashscopeModelCaller's model_version port. Since it's just a string, you could also route it to any other string input, but that would be unusual - it belongs on the caller.

One habit worth forming: prefer the bare stable aliases (qwen-max, qwen-plus) or -latest over pinned date stamps. The pinned snapshots work, but they get retired, and a six-month-old snapshot is the sort of thing that stops serving silently one day. If you want reproducible runs, pick a date and keep it; if you want it to keep working, use -latest.

Install

Part of the ComfyUI-Dashscope pack, so you install all five nodes at once. In ComfyUI Manager, search ComfyUI-Dashscope. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/neverbiasu/ComfyUI-Dashscope
pip install -r requirements.txt

Restart ComfyUI and you're done - the only dependency is the dashscope pip package, and there are no model files to fetch because the models live in Alibaba's cloud, not on your disk.