Qwen (Alibaba)

Qwen is Alibaba's open-weight model family, including the Qwen3-Coder series optimized for software development. AgentXchain connects via api_proxy using Alibaba's DashScope API or locally via Ollama.

Which adapter?

api_proxy — either via DashScope cloud API (provider: "openai" with base_url) or locally via Ollama.

Cloud API (DashScope)

Prerequisites

• A DashScope API key — get one from dashscope.aliyuncs.com
• DASHSCOPE_API_KEY set in your environment

Configuration

{
  "runtimes": {
    "qwen-dev": {
      "type": "api_proxy",
      "provider": "openai",
      "model": "qwen3-coder-480b-a35b",
      "auth_env": "DASHSCOPE_API_KEY",
      "base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions"
    }
  },
  "roles": {
    "dev": {
      "runtime": "qwen-dev",
      "mandate": "Implement features and fix bugs",
      "authority": "proposed"
    }
  }
}

Available models

Model	Params	Best for
qwen3-coder-480b-a35b	480B (35B active)	Most capable, MoE architecture
qwen3-coder-30b-a3b	30B (3B active)	Efficient local deployment
qwen3-coder-next	—	Latest experimental
qwen3.6-plus	—	General-purpose reasoning

Local via Ollama

Qwen's smaller coder models are excellent for local use:

ollama pull qwen3-coder:30b

{
  "runtimes": {
    "qwen-local": {
      "type": "api_proxy",
      "provider": "ollama",
      "model": "qwen3-coder:30b",
      "auth_env": "OLLAMA_API_KEY"
    }
  }
}

Verify the connection

export DASHSCOPE_API_KEY="sk-..."
agentxchain connector check

Gotchas

• MoE architecture: The 480B model uses Mixture of Experts (only 35B active parameters per token), making it efficient despite the large total parameter count.
• DashScope region: The API endpoint is hosted in China. Latency may be higher from other regions.
• Local deployment: The 30B-A3B variant is one of the best coding models for local deployment via Ollama, requiring only ~8GB of memory with 4-bit quantization.