Blitz

Model Selection

Legion CLI supports 500+ AI models with intelligent selection based on task requirements.

Model Categories

Flagship Models

Best for complex coding tasks, architecture, and detailed analysis:

Model	Provider	Strengths
Claude 3.5 Sonnet	Anthropic	Coding, analysis, long context
GPT-4o	OpenAI	General purpose, fast
Gemini 1.5 Pro	Google	Long context, multimodal
Claude 3 Opus	Anthropic	Reasoning, complex tasks

Fast Models

Best for simple tasks, titles, and quick operations:

Model	Provider	Strengths
Claude 3.5 Haiku	Anthropic	Fast, cheap
GPT-4o-mini	OpenAI	Fast, cost-effective
Gemini 1.5 Flash	Google	Ultra-fast

Reasoning Models

Best for math, logic, and complex reasoning:

Model	Provider	Strengths
o1	OpenAI	Step-by-step reasoning
Claude 3 Opus	Anthropic	Deep analysis

Selection Criteria

By Task Type

Task	Recommended Model
Code generation	Claude 3.5 Sonnet, GPT-4o
Code review	Claude 3.5 Sonnet
Bug fixing	Claude 3.5 Sonnet, GPT-4o
Refactoring	Claude 3.5 Sonnet
Documentation	Claude 3.5 Haiku, GPT-4o-mini
Testing	Claude 3.5 Haiku
Architecture	Claude 3 Opus, o1
Quick edits	Claude 3.5 Haiku

By Context Size

Context	Models
Small (< 10K tokens)	Any model
Medium (10K-100K)	Claude 3.5, GPT-4o
Large (100K-200K)	Claude 3.5, Gemini 1.5
Huge (200K+)	Gemini 1.5 (1M context)

By Cost

Cost	Models
Low	Haiku, GPT-4o-mini, Flash
Medium	Claude 3.5 Sonnet, GPT-4o
High	Claude 3 Opus, o1

Configuration

Primary Model

{
  "model": "anthropic/claude-sonnet-4"
}

Small Model

For simple tasks like title generation:

{
  "small_model": "anthropic/claude-haiku-4"
}

Subagent Model

For background tasks:

{
  "subagent_model": "anthropic/claude-sonnet-4"
}

Model Switching

Runtime Switching

# Use specific model
legion run "implement feature" --model openai/gpt-4o

# Use small model for quick task
legion run "generate title" --model anthropic/claude-haiku-4

Automatic Selection

Legion automatically selects models based on:

• Task complexity
• Required capabilities
• Cost constraints
• Rate limits

Model Capabilities

Vision

Models that support image input:

• Claude 3.5 Sonnet
• GPT-4o
• Gemini 1.5 Pro

Tool Use

Models that support function calling:

• Claude 3.5 Sonnet
• GPT-4o
• Gemini 1.5 Pro

Streaming

All modern models support streaming.

JSON Mode

Models with reliable JSON output:

• GPT-4o
• Claude 3.5 Sonnet
• Gemini 1.5 Pro

Custom Models

OpenAI-Compatible

{
  "providers": {
    "custom": {
      "type": "openai",
      "baseUrl": "https://your-api.com/v1",
      "apiKey": "env:CUSTOM_API_KEY",
      "models": ["model-1", "model-2"]
    }
  }
}

Model Registry

{
  "models": {
    "custom-model": {
      "provider": "custom",
      "id": "model-1",
      "name": "Custom Model",
      "capabilities": ["tools", "streaming"],
      "contextWindow": 128000
    }
  }
}

Cost Optimization

Token Usage

Track token usage:

legion usage show
legion usage show --model anthropic/claude-sonnet-4

Cost Limits

Set cost limits:

{
  "providers": {
    "anthropic": {
      "costLimit": {
        "daily": 10.00,
        "monthly": 100.00
      }
    }
  }
}

Model Fallbacks

Configure fallbacks for cost optimization:

{
  "model": "anthropic/claude-sonnet-4",
  "fallback": {
    "model": "anthropic/claude-haiku-4",
    "onCostLimit": true
  }
}

Best Practices

1. Start with Claude 3.5 Sonnet - Best for most coding tasks
2. Use Haiku for simple tasks - Title generation, quick edits
3. Use Opus for complex reasoning - Architecture, design
4. Monitor costs - Use cost limits and tracking
5. Set up fallbacks - Handle rate limits and errors
6. Test different models - Find what works for your workflow

This site is open source. Improve this page.