Model Strategy

Overview

Automatically select optimal Claude model based on task complexity to balance performance and cost. Use higher reasoning models for complex tasks, faster models for simple tasks.

When to Use

Apply this strategy BEFORE starting any task involving:

• Code analysis or architecture review
• Writing or refactoring
• Technical research or exploration
• Complex problem-solving

Match task type to model:

• Deep Analysis: Complex reasoning, architecture patterns, system design
• Assembly/Format/Search: Simple file operations, formatting, basic search
• Extreme Complexity: Formal verification, consensus algorithms, advanced optimization

Core Strategy

Complexity-Based Model Selection

| Task Type | Model | Use When | |-----------|-------|----------| | Extreme Complexity | opus-4-6 | Formal verification, distributed consensus, Byzantine fault tolerance, proof systems | | Deep Analysis | sonnet-4-6 | Architecture review, system design, complex code analysis, multi-service coordination | | Assembly/Format/Search | haiku-4-5 | File search, formatting, simple operations, list generation, basic transformations |

Decision Tree

Is task extremely complex?
├─ YES → Use opus-4-6
└─ NO → Is it deep analysis/architecture/system design?
    ├─ YES → Use sonnet-4-6
    └─ NO → Use haiku-4-5 (assembly, format, search)

Examples

✅ Correct Application

Task: Analyze microservices architecture with 15 services

Decision: Deep analysis → sonnet-4-6
Reasoning: Requires architectural pattern recognition, service interaction analysis

Task: Search 100 files and format list

Decision: Simple search/format → haiku-4-5
Reasoning: Basic file operations, no complex reasoning needed

Task: Design distributed consensus algorithm

Decision: Extreme complexity → opus-4-6
Reasoning: Formal verification, Byzantine failures, 1000+ node optimization

Quick Reference

# Decision workflow:
# 1. Can you solve this in 5 minutes with simple commands? → haiku-4-5
# 2. Does this require architectural thinking? → sonnet-4-6
# 3. Does this require formal proofs or consensus protocols? → opus-4-6

Common Patterns:

• Code formatting, file search → haiku-4-5
• Refactoring, pattern analysis → sonnet-4-6
• Algorithm design, formal verification → opus-4-6

Common Mistakes

❌ Using opus-4-6 for simple tasks

• Wastes resources on basic operations
• haiku-4-5 is faster and cheaper for simple tasks

❌ Using haiku-4-5 for complex analysis

• May miss architectural patterns
• sonnet-4-6 designed for deep analysis

❌ Not stating model choice upfront

• Makes reasoning unclear
• Add model decision to your opening statement

Alternative Models (Fallbacks)

If the preferred Claude model is unavailable, use these fallbacks based on provider:

Extreme Complexity (Formal verification, distributed systems, advanced proofs)

• Primary: anthropic/claude-opus-4-6 (Anthropic)
• Fallbacks:
  • openrouter/openai/gpt-4o (OpenAI)
  • openrouter/google/gemini-ultra (Google)
  • openrouter/anthropic/claude-3-opus (Anthropic legacy)

Deep Analysis (Architecture review, system design, complex reasoning)

• Primary: anthropic/claude-sonnet-4-6 (Anthropic)
• Fallbacks:
  • openrouter/openai/gpt-4 (OpenAI)
  • openrouter/anthropic/claude-3-sonnet (Anthropic legacy)
  • openrouter/google/gemini-pro (Google)
  • openrouter/meta-ai/llama-3.1-405b (Meta)

Assembly/Format/Search (Simple tasks, formatting, basic transformations)

• Primary: anthropic/claude-haiku-4-5 (Anthropic)
• Fallbacks:
  • openrouter/stepfun/step-3.5-flash (StepFun)
  • openrouter/openai/gpt-4o-mini (OpenAI)
  • openrouter/anthropic/claude-3-haiku (Anthropic legacy)
  • openrouter/meta-ai/llama-3.1-70b (Meta)

Note: When using OpenRouter models, prefix with openrouter/ as shown. For direct provider access (if configured), use the appropriate model ID.

Implementation

Always state your model choice when starting a task:

I'll use [model] because [task type].

Example:

I'll use sonnet-4-6 because this requires deep analysis of the
microservices architecture pattern.