Model Routing — Intelligent Task-Based Model Selection

Model Routing — Intelligent Task-Based Model Selection

This skill guides Claude in automatically selecting the most appropriate model based on task complexity and nature, balancing quality, speed, and cost.

Routing Decision Matrix

🟢 Haiku — Fast & Lightweight

Use when the task involves:

• Simple questions or factual lookups
• File search, grep, directory listing
• Quick text formatting or transformations
• Generating short boilerplate code
• Answering single-step, well-defined questions
• Summarizing short content (< 500 tokens)
• Log parsing or pattern matching

Signals: Task can be answered in one step, no reasoning chain required, low stakes, high frequency.

---

🟡 Sonnet — Balanced & General-Purpose (default)

Use when the task involves:

• General software development tasks
• Implementing features or fixing bugs
• Code review and refactoring
• Writing tests for existing code
• PR review and description generation
• Multi-file edits with clear scope
• Moderate debugging (known error class)
• Writing documentation or READMEs
• Explaining or analyzing code

Signals: Task requires moderate reasoning, 2–5 step execution, standard engineering work.

---

🔴 Opus — Deep Reasoning & Critical Work

Use when the task involves:

• System or architecture design decisions
• Complex debugging with unknown root cause
• Security analysis and threat modeling
• Evaluating trade-offs between multiple approaches
• Generating or reviewing critical business logic
• Migrating large codebases or breaking changes
• Tasks where errors have high impact or are hard to reverse
• Cross-cutting concerns affecting multiple systems
• Performance optimization requiring deep profiling

Signals: Multi-step reasoning required, high stakes, ambiguity present, architectural impact.

---

How to Apply This Skill

When starting a task, evaluate:

1. Scope: Is this one step or many?
2. Stakes: What's the cost of a mistake?
3. Reasoning depth: Does this need chain-of-thought or inference?
4. Frequency: Is this a repeated lightweight task?

Then select accordingly and, when relevant, explain the routing choice to the user:

"This is an architectural decision with broad impact — I'll use Opus for deeper analysis." "This is a quick file search — Haiku is the right fit here."

Important Notes

• Default to Sonnet when the complexity is unclear.
• Never downgrade critical tasks to save cost — correctness comes first.
• Prefer Haiku for agentic loops where the same operation repeats many times.
• If the user has explicitly set a model preference in this session, respect it and do not override.