pi-provider-alibaba-universal

Alibaba Cloud Model Studio providers for Pi.

Six providers covering every Alibaba plan — Coding Plan, Token Plan, and Pay-as-you-go — across both international and Chinese endpoints. All chat models appear with correct reasoning levels, context windows, and input types without extra configuration.

Providers

| Provider ID | Plan | Endpoint | |---|---|---| | alibaba-coding-intl | Coding Plan (Intl) | coding-intl.dashscope.aliyuncs.com | | alibaba-coding-cn | Coding Plan (CN) | coding.dashscope.aliyuncs.com | | alibaba-token-intl | Token Plan (Intl) | token-plan.ap-southeast-1.maas.aliyuncs.com | | alibaba-token-cn | Token Plan (CN) | token-plan.cn-beijing.maas.aliyuncs.com | | alibaba-api-intl | DashScope API (Intl) | dashscope-intl.aliyuncs.com | | alibaba-api-cn | DashScope API (CN) | dashscope.aliyuncs.com |

Install

Do not install this package together with pi-provider-alibaba or pi-provider-alibaba-complete: all three register the same provider IDs.

pi install npm:pi-provider-alibaba-universal

Or try it for a single session:

pi -e npm:pi-provider-alibaba-universal

Configure

Set the credential for the product you purchased, restart Pi, then select the matching provider and model:

# Token Plan
export ALIBABA_TOKEN_PLAN_API_KEY="<your-token-plan-key>"

# Coding Plan
export ALIBABA_CODING_API_KEY="<your-coding-plan-key>"

# Pay-as-you-go DashScope API
export ALIBABA_API_KEY="<your-api-key>"

China and international routes use different endpoints. Select the route matching your key's product and region.

Token Plan models

The Token Plan providers ship with a complete static catalog of all 12 chat-capable models documented by Alibaba, available immediately without waiting for API discovery:

| Model | Reasoning | Input | Context | Max output | |---|---|---|---|---| | auto | — (router) | text | 128K | 8K | | qwen3.8-max | low · medium · xhigh | text · image | 1M | 128K | | qwen3.8-flash | off · high | text · image | 1M | 64K | | qwen3.7-max | off · high | text | 1M | 64K | | qwen3.7-plus | off · high | text · image | 1M | 64K | | qwen3.6-flash | off · high | text · image | 1M | 64K | | deepseek-v4.1-flash | off · high · max | text · image | 1M | 384K | | deepseek-v4-pro | off · high · max | text | 1M | 384K | | deepseek-v4-pro-0813 | off · low · high · max | text | 1M | 384K | | deepseek-v4-flash-0731 | off · low · high · max | text | 1M | 384K | | glm-5.3 | off · high · max | text | 1M | 128K | | glm-5.2 | off · high · max | text | 1M | 128K |

Multimodal generation models (qwen-image-3.0-pro, qwen-audio-*, wan2.7-*, happyhorse-*) are excluded because they use dedicated non-chat endpoints. See Alibaba's multimodal integration guide.

Coding Plan and API models

Coding Plan and DashScope API providers use dynamic model discovery via the /v1/models endpoint. Open /model in Pi to refresh the list. Models appear with correct capability metadata even when discovered dynamically.

Reasoning controls

| Model family | Selectable levels | Wire format | |---|---|---| | DeepSeek V4 (current) | off · high · max | enable_thinking + reasoning_effort | | DeepSeek V4 (dated 0813/0731) | off · low · high · max | enable_thinking + reasoning_effort | | Qwen 3.8 Max | low · medium · xhigh | enable_thinking + reasoning_effort | | Qwen 3.6 / 3.7 / 3.8 Flash | off · high | enable_thinking only | | GLM-5 | off · high · max | enable_thinking + reasoning_effort |

All providers use system role, max_tokens field, and Alibaba's enable_thinking wire format. Unknown model families stay conservative until their protocol is confirmed.

Model refresh

Pi refreshes the model list each time you open /model:

• Token Plan: the static catalog is always present. Dynamic discovery adds any newly released models without replacing existing entries.
• Coding Plan / API: dynamic discovery replaces the current list with fresh results from the endpoint.

Verification

npm install
npm run verify

Runs typecheck, 38 tests, and packaging dry-run.

License

MIT

Acknowledgments

This package is based on the provider architecture and capability projection work from:

• dinhe/pi-provider-alibaba — the original Alibaba Cloud provider for Pi, which established the route, transport, and authentication layer for all six Alibaba endpoints.
• weijiafu14/pi-provider-alibaba-complete — which added the first reasoning capability projection, model-specific compatibility switches, and wire-format verification.

This package extends that foundation with a static-first model catalog for Token Plan, expanded capability mappings for newer models, and a dynamic merge strategy that preserves curated metadata across refreshes.