Module 3: Model SelectionLesson 3 of 4

The Starter Stack

The Starter Stack

Here's exactly what models to use when starting out.

The Three-Model Setup

models: # Your main agent — needs to be smart default: anthropic/claude-sonnet-4 # Aliases for easy switching aliases: opus: anthropic/claude-opus-4 # When you need power sonnet: anthropic/claude-sonnet-4 # Default workhorse haiku: anthropic/claude-haiku # Quick/cheap tasks

When to Use Each

Opus (The Thinker)

Use for: - Your main interactive session - Complex planning - Important decisions - Multi-step reasoning Don't use for: - Background tasks - Simple lookups - High-volume work

Sonnet (The Workhorse)

Use for: - Cron jobs - Sub-agents - General tasks - Writing/coding Don't use for: - Tasks requiring deep reasoning - When quality is critical

Haiku (The Sprinter)

Use for: - Quick data extraction - Simple transformations - High-volume processing - When speed > quality Don't use for: - Anything requiring nuance - Complex instructions

Real Cost Comparison

Processing 1 million tokens:

ModelInputOutputTotal
Opus$15$75$90
Sonnet$3$15$18
Haiku$0.25$1.25$1.50

That's a 60x cost difference between Opus and Haiku!

My Actual Setup

# Main session — I'm talking to this, needs to be good default: anthropic/claude-opus-4 # Cron jobs — background, don't need Opus cron: model: anthropic/claude-sonnet-4 # Sub-agents — focused tasks subagent: default: anthropic/claude-sonnet-4

Monthly cost: ~$50-100 with heavy use

If I used Opus for everything: ~$300+

The Progression

  1. Start with Sonnet everywhere (learn the system)
  2. Upgrade main session to Opus (when you want more)
  3. Add Haiku for specific fast tasks (optimize costs)

Don't over-optimize early. Get it working first, then tune.