--model Option

Specify the AI model to use for the session

Overview

The --model (or -m) option allows you to specify which AI model Qwen Code should use for the session. Different models offer different capabilities, speeds, and costs.

Syntax

qwen --model <model-name>
qwen -m <model-name>

Available Models

Qwen Models (Dashscope)

qwen-coder-plus

Best for complex coding tasks

qwen --model qwen-coder-plus

Specifications:

• Context: 256K tokens
• Quality: Highest
• Speed: Medium
• Cost: $$$

Use for:

• Multi-file refactoring
• Complex algorithms
• Architecture decisions
• Large codebase navigation

qwen-max

Balanced performance

qwen --model qwen-max

Specifications:

• Context: 200K tokens
• Quality: High
• Speed: Medium
• Cost: $$

Use for:

• General development
• Feature implementation
• Code reviews
• Documentation

qwen-turbo

Fastest responses

qwen --model qwen-turbo

Specifications:

• Context: 128K tokens
• Quality: Good
• Speed: Fast
• Cost: $

Use for:

• Quick questions
• Simple code generation
• Explanations
• Rapid iteration

OpenAI Models

When using OpenAI authentication:

# GPT-4 Turbo
qwen --model gpt-4-turbo --auth-type openai

# GPT-4
qwen --model gpt-4 --auth-type openai

# GPT-3.5 Turbo
qwen --model gpt-3.5-turbo --auth-type openai

Anthropic Models

When using Anthropic authentication:

# Claude 3 Opus (most capable)
qwen --model claude-3-opus-20240229 --auth-type anthropic

# Claude 3 Sonnet (balanced)
qwen --model claude-3-sonnet-20240229 --auth-type anthropic

# Claude 3 Haiku (fastest)
qwen --model claude-3-haiku-20240307 --auth-type anthropic

Basic Usage

Specify Model at Launch

qwen --model qwen-coder-plus

With Other Options

# Headless mode with specific model
qwen --model qwen-turbo --prompt "Quick question"

# YOLO mode with powerful model
qwen --model qwen-coder-plus --yolo --prompt "Refactor entire project"

# JSON output with balanced model
qwen --model qwen-max --prompt "Generate code" --output-format json

Model Selection Strategy

For Development Tasks

# Complex refactoring
qwen --model qwen-coder-plus --prompt "Refactor authentication system"

# Quick fixes
qwen --model qwen-turbo --prompt "Fix this type error"

# Code review
qwen --model qwen-max --prompt "Review these changes"

For Different Project Sizes

# Large monorepo (needs large context)
cd large-project
qwen --model qwen-coder-plus

# Small utility (fast model is fine)
cd small-util
qwen --model qwen-turbo

# Medium project (balanced)
cd medium-app
qwen --model qwen-max

Model Comparison

Model	Context	Speed	Quality	Cost	Best For
qwen-coder-plus	256K	⭐⭐	⭐⭐⭐	$$$	Complex coding
qwen-max	200K	⭐⭐	⭐⭐	$$	General purpose
qwen-turbo	128K	⭐⭐⭐	⭐⭐	$	Quick tasks
gpt-4-turbo	128K	⭐⭐	⭐⭐⭐	$$$$	Advanced reasoning
claude-3-opus	200K	⭐⭐	⭐⭐⭐	$$$$	Complex analysis
claude-3-haiku	200K	⭐⭐⭐	⭐⭐	$	Fast responses

Configuration

Via Settings File

Set a default model:

// settings.json
{
  "ai": {
    "model": "qwen-coder-plus",
    "provider": "dashscope"
  }
}

Via Environment Variable

export QWEN_MODEL="qwen-coder-plus"
qwen

Project-Specific Defaults

// .qwen/settings.json
{
  "ai": {
    "model": "qwen-max",  // Override global default
    "provider": "dashscope"
  }
}

Precedence Order

1. Command-line --model flag (highest priority)
2. Environment variable QWEN_MODEL
3. Project settings .qwen/settings.json
4. Global settings ~/.qwen/settings.json (lowest priority)

Advanced Usage

Switch Models Mid-Session

In interactive mode:

qwen --model qwen-turbo
> Ask quick questions
> /model  # Switch to qwen-coder-plus
> Now do complex refactoring

Model-Specific Prompts

# Use fast model for exploration
qwen --model qwen-turbo --prompt "Explore the codebase and summarize"

# Then use powerful model for implementation
qwen --model qwen-coder-plus --continue --prompt "Implement the feature"

Cost Optimization

#!/bin/bash
# Use appropriate model based on task complexity

TASK_COMPLEXITY=$1
PROMPT=$2

if [ "$TASK_COMPLEXITY" = "simple" ]; then
  MODEL="qwen-turbo"
elif [ "$TASK_COMPLEXITY" = "complex" ]; then
  MODEL="qwen-coder-plus"
else
  MODEL="qwen-max"
fi

qwen --model "$MODEL" --prompt "$PROMPT"

Custom Models

OpenAI-Compatible APIs

Use custom or self-hosted models:

qwen --model custom-model \
     --openai-base-url https://your-api.com/v1 \
     --openai-api-key your-key

Configuration for Custom Models

{
  "ai": {
    "provider": "openai",
    "model": "custom-model",
    "baseUrl": "https://your-api.com/v1",
    "apiKey": "${CUSTOM_API_KEY}"
  }
}

Model Capabilities

Tool Support

All models support Qwen Code tools:

✅ File operations (read, write, edit)
✅ Shell commands
✅ Code search
✅ Git operations
✅ Web search

Feature Comparison

Feature	qwen-coder-plus	qwen-max	qwen-turbo
Multi-file edits	Excellent	Good	Basic
Code generation	Excellent	Good	Good
Explanations	Excellent	Excellent	Good
Debugging	Excellent	Good	Basic
Speed	Medium	Medium	Fast
Context size	256K	200K	128K

Real-World Examples

Web Development

# Frontend (fast iteration)
qwen --model qwen-turbo --prompt "Add a button to the navbar"

# Backend (complex logic)
qwen --model qwen-coder-plus --prompt "Implement OAuth flow"

# Full-stack (balanced)
qwen --model qwen-max --prompt "Build user profile page"

Data Science

# Quick analysis
qwen --model qwen-turbo --prompt "Plot this data"

# Complex ML model
qwen --model qwen-coder-plus --prompt "Build gradient boosting model"

DevOps

# Simple scripts
qwen --model qwen-turbo --prompt "Write deploy script"

# Complex infrastructure
qwen --model qwen-coder-plus --prompt "Design Kubernetes architecture"

Performance Tips

Match Model to Task

# Don't use powerful model for simple tasks
# Bad:
qwen --model qwen-coder-plus --prompt "What's 2+2?"

# Good:
qwen --model qwen-turbo --prompt "What's 2+2?"

Use Fast Models for Iteration

# Iterate quickly with turbo
for i in {1..5}; do
  qwen --model qwen-turbo --prompt "Try approach $i"
done

# Then implement with coder-plus
qwen --model qwen-coder-plus --prompt "Implement best approach"

Monitor Token Usage

# Check context usage
qwen --model qwen-coder-plus --prompt "Task" --output-format json | \
  jq '.usage.totalTokens'

# Switch model if approaching limit
if [ $TOKENS -gt 200000 ]; then
  # Context nearly full, compress or switch model
  qwen --prompt "/compress"
fi

Troubleshooting

Model Not Available

Error: Model 'qwen-coder-plus' is not available

Solutions:

1. Check authentication:

   qwen --auth-type dashscope --dashscope-api-key "$KEY"
   

2. Verify model name:

   # List available models
   qwen --model invalid 2>&1 | grep "Available models"
   

3. Check API access:

   # Test with curl
   curl -H "Authorization: Bearer $DASHSCOPE_API_KEY" \
        https://dashscope.aliyuncs.com/api/v1/models
   

Authentication Issues

Error: Authentication type not available

Solution:

# Set auth type matching the model
qwen --model qwen-coder-plus --auth-type dashscope

# Or for OpenAI models
qwen --model gpt-4 --auth-type openai

Invalid Model Name

Error: Unknown model: 'qwen-coder'

Correct names:

• ✅ qwen-coder-plus (not qwen-coder)
• ✅ qwen-turbo (not qwen-fast)
• ✅ qwen-max (not qwen-large)

Best Practices

Start Fast, Scale Up

Begin with fast models, upgrade as needed:

# Explore with turbo
qwen --model qwen-turbo
> Understand the codebase

# Switch to coder-plus for implementation
> /model
# Select qwen-coder-plus
> Implement the feature

Project Defaults

Set appropriate defaults per project:

// Large enterprise app
{
  "ai": {
    "model": "qwen-coder-plus"  // Need large context
  }
}

// Small utility
{
  "ai": {
    "model": "qwen-turbo"  // Speed matters more
  }
}

Cost Awareness

Monitor costs and adjust:

# Track usage
qwen --prompt "Task" --output-format json | \
  jq '.stats.estimatedCost'

# Switch to cheaper model if budget-conscious
qwen --model qwen-turbo

Context Window Management

Choose models with appropriate context:

# Large codebase, many files
qwen --model qwen-coder-plus  # 256K context

# Small, focused task
qwen --model qwen-turbo  # 128K sufficient

See Also

/model Command

Switch models in interactive mode

Authentication

Set up model provider authentication

Model Comparison

Detailed model benchmarks and comparison

Configuration

Configure default models and settings