Coding Agent setup

Qwen Code setup: API key, Base URL, and models

Configure Qwen Code with a BetterToken API Key, Base URL, and Responses model settings. Review the tested limitations for file-reading tasks.

Connect Qwen Code to BetterToken through modelProviders.openai in ~/.qwen/settings.json. In macOS testing with a GPT model, Qwen Code 0.26.0 completed text requests with both Responses and Chat Completions.

Key settings

Field	Value
API Key	BetterToken API Key, referenced by `envKey`
Base URL	`https://bettertoken.ai/v1` in `baseUrl`
Model	`YOUR_MODEL_ID`
Provider configuration	`modelProviders.openai`
API format	`wireApi: "responses"`

Prerequisites

Node.js 22 or later for the npm installation below
A BetterToken API Key: register and get one
A Model ID copied from the model plaza

Install

Use the npm option from the official quickstart:

npm install -g @qwen-code/qwen-code@latest
qwen --version

Manual setup

Set the API Key in your terminal, replacing YOUR_API_KEY:

export BETTERTOKEN_API_KEY="YOUR_API_KEY"

Merge these settings into ~/.qwen/settings.json. Replace both YOUR_MODEL_ID values with the same exact Model ID:

{
  "modelProviders": {
    "openai": [
      {
        "id": "YOUR_MODEL_ID",
        "envKey": "BETTERTOKEN_API_KEY",
        "baseUrl": "https://bettertoken.ai/v1",
        "wireApi": "responses"
      }
    ]
  },
  "security": {
    "auth": {
      "selectedType": "openai"
    }
  },
  "model": {
    "name": "YOUR_MODEL_ID"
  }
}

envKey names an environment variable; it does not contain the API Key itself. The official model provider reference also supports wireApi: "chat-completions". Responses had more successful file-reading attempts in this test, but neither format passed every attempt.

Verify the connection

Run this text-only check in the same terminal:

qwen -p "Reply with only: OK" --output-format stream-json --max-session-turns 4

Check the final answer for OK. A successful text request does not establish that a file-reading task will produce a correct final answer; use the limitation above when choosing your workflow.

Switch models

Add a model entry under modelProviders.openai, set model.name to its exact id, and start a new session.

Common errors

Symptom	What to check
Authentication fails	Export the variable named by `envKey` in the terminal running Qwen Code.
Wrong endpoint	Use `https://bettertoken.ai/v1` in `baseUrl`; `wireApi` selects the request protocol.
Model is not found	Make `model.name` match an `id` under `modelProviders.openai`.
Empty or incorrect answer after a file read	This occurred even with correct tool results in 0.26.0. Switching to Responses did not eliminate the failures; use another tested client for file-reading tasks.

Technical details

wireApi: "responses" selects /v1/responses; wireApi: "chat-completions" selects /v1/chat/completions. Both text paths passed in Qwen Code 0.26.0 testing. File-reading outcomes were 2/3 and 0/3 respectively.