pi-provider-alibaba-bailian

A pi provider extension for Alibaba Bailian (DashScope, OpenAI-compatible mode).

Setup

Set your DashScope API key as an environment variable:

export ALIBABA_API_KEY=sk-...

Install

pi install npm:pi-provider-alibaba-bailian

Then in pi: run /reload (or restart pi), then pick a model with /model → alibaba-bailian/<model>.

Endpoint

https://dashscope.aliyuncs.com/compatible-mode/v1 — the public DashScope OpenAI-compatible endpoint. Any DashScope API key works here.

Available models

| Model ID | Input | Context | Output | Notes | | --- | --- | --- | --- | --- | | qwen3.8-max | text+image | 1M | 131k | qwen toggle + reasoning_effort (low/med/xhigh) | | qwen3.7-plus | text+image | 1M | 64k | thinking toggle (qwen) | | glm-5.2 | text | 1M | 131k | reasoning_content auto-captured, no toggle | | kimi-k2.7-code | text+image | 262k | 262k | thinking always-on, no disable (qwen) | | kimi/kimi-k3 | text+image | 1M | 131k | thinking always-on, no disable (qwen) | | kimi-k2.6 | text+image | 262k | 16k | thinking toggle (qwen) | | deepseek-v4-pro | text | 1M | 384k | effort-based thinking (deepseek) | | deepseek-v4-flash | text | 1M | 384k | effort-based thinking (deepseek) | | MiniMax/MiniMax-M3 | text+image | 512k | 128k | no toggle, reasoning captured |

Thinking levels are selectable per request and follow each model's real capabilities on Bailian: qwen3.7-plus / kimi-k2.6 expose off + high (enable_thinking toggle); qwen3.8-max exposes off + low/medium/xhigh (enable_thinking + reasoning_effort); deepseek-v4 exposes off + high + max (reasoning_effort); thinking-only models (kimi-k2.7-code, kimi-k3, glm-5.2, MiniMax-M3) cannot disable thinking and expose high only. Explicit prompt caching (Anthropic-style cache_control) is enabled for all models. Pricing is sourced from the models.dev database (with OpenRouter fallback for some models).

How it works

pi.registerProvider("alibaba-bailian", { ... }) registers the provider with its endpoint, auth ($ALIBABA_API_KEY env var), and a per-model catalog (context window, output limit, pricing, reasoning/compat options). The extension only runs inside pi — list it as a peerDependency on @earendil-works/pi-coding-agent.

License

MIT