Qwen Token Plan

Rho supports QwenCloud Token Plan through its OpenAI-compatible Chat Completions API.

Provider details

Setting	Value
Provider	qwen-token-plan
Auth	qwen-token-plan-api-key
Environment override	QWEN_TOKEN_PLAN_API_KEY
API base	https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1
Model source	Token Plan /models API

Create a Token Plan dedicated API key on the API Keys page. Token Plan keys are not interchangeable with Coding Plan or pay-as-you-go keys. Model access depends on your plan, so Rho fetches the model list after login instead of keeping a static list.

Interactive login

In the TUI, run:

/login qwen-token-plan

Rho asks for your Token Plan API key, stores it in the configured credential store, and refreshes the models available to your account. Select one with /model, for example:

/model qwen-token-plan/qwen3.7-plus

Remove the stored key with:

/logout qwen-token-plan

Environment and automation

For CI or development, set QWEN_TOKEN_PLAN_API_KEY. It overrides a stored key.

export QWEN_TOKEN_PLAN_API_KEY="<api-key>"
rho --provider qwen-token-plan \
  --auth qwen-token-plan-api-key \
  --model qwen-token-plan/qwen3.7-plus \
  run "review this project"

Rho sends the key as a Bearer token. Do not put the key in config.toml or commit it to source control.

Models and reasoning

Use /config and choose Refresh model lists to fetch the current models for your Token Plan account. Rho maps models.dev reasoning metadata for the alibaba-token-plan catalog:

• qwen3.8-max and qwen3.8-max-preview: low, medium, and xhigh (API default is xhigh when the field is omitted)
• Models that only advertise a thinking toggle may not expose a full effort cycle until models.dev lists exact effort values

When thinking models stream reasoning_content alongside tool calls, Rho retains that field as exact-model provider context and replays it on later assistant messages in the same tool loop. Without replay, multi-step tool use can degrade into empty assistant responses after a couple of tool rounds. Rho orchestration always uses the streaming path for this; non-stream send_turn / complete responses cannot carry provider context.

See Token Plan overview and the Token Plan FAQ for quota, region, and key rules.

Note: QwenCloud Coding Plan is a separate product (different keys, base URL, and model allowlist). Rho's qwen-token-plan provider targets Token Plan endpoints only.