The Strategic Model Selection Matrix
Sophistication Required
Opus
High Sophistication, Low Iteration
Complex reasoning in focused bursts
Strategic insights, board Q&A prep, competitive analysis
Opus then Sonnet
High Sophistication, High Iteration
Start with Opus for framework, continue with Sonnet
Board presentation: Opus for narrative arc, Sonnet for slide development
Haiku
Low Sophistication, Low Iteration
Fast execution with more capability than you expect
Email drafting, classification, structured extraction, quick Q&A, data formatting
Sonnet
Low Sophistication, High Iteration
Most executive work lives here: iterative refinement
Proposal development, report writing, presentation creation
Low iteration
Iteration Needs →
High iteration
Real-World Scenarios
Quarterly Board Presentation
1
Strategic Framework
Define narrative arc, key messages, strategic positioning
Opus2-3 exchanges
2
Slide Development
Create, refine, and iterate on presentation content
Sonnet8-12 exchanges
3
Q&A Preparation
Anticipate tough questions, prepare nuanced responses
Opus1-2 per question
4
Final Formatting
Polish slides, check consistency, format speaker notes
Haiku2-3 exchanges
Strategic Client Proposal
1
Understand Client Needs
Analyze requirements, identify strategic opportunities
Opus2-3 exchanges
2
Develop Proposal
Write sections, refine value proposition, iterate on approach
Sonnet6-10 exchanges
3
Polish and Format
Final edits, formatting, executive summary
Haiku2-3 exchanges
Common misconception
Executive Summaries Are Not Simple Tasks
Executive summaries require strategic synthesis, not just compression. They need to identify key insights, frame strategic context, and guide decision-making. Sonnet handles this well. Haiku does not, despite the temptation to treat summaries as "simple."
Why Executives Get Model Selection Wrong
×
Conflating importance with sophistication: Just because a task is important to your business does not mean it requires complex reasoning. An important email might only need Haiku's clear execution.
×
Defaulting to Opus out of habit: The gap between Sonnet and Opus has narrowed dramatically. Sonnet 4.6 now handles work that genuinely required Opus six months ago. Use Opus for its specific advantages: Agent Teams, 128K output ceiling, and first-attempt correctness on ambiguous multi-step problems.
×
Underestimating Haiku: Haiku 4.5 rivals earlier Sonnet models in capability. Using Sonnet for classification, extraction, and simple Q&A wastes resources when Haiku handles these tasks well at a fraction of the cost.
×
Not switching models mid-project: Staying in one model for an entire project instead of using each model for its strength. Think relay race, not marathon.
Strategic Principles for Model Selection
Match Model to Cognitive Work Type
The importance of a task does not determine the model. The type of thinking required does. Board presentations need all three models at different stages. Routine emails need only Haiku, regardless of recipient importance.
Context is Currency: Spend Wisely
Every exchange uses context window "budget," though that budget has grown dramatically. Both Opus and Sonnet now support 1M token context windows, and automatic compaction can extend conversations further. The principle still holds: Opus burns through resources fastest with deeper responses. But the constraint has shifted from "you will run out of context" to "you are paying for capability you may not need."
Most Executives Over-Sophisticate
70% of executive work requires iterative refinement (Sonnet), not complex reasoning (Opus). This was true in November 2025, and the case has only gotten stronger: Sonnet 4.6 now performs within 1-2 points of Opus on most benchmarks. We overestimate the sophistication needed because we conflate business importance with cognitive complexity. Start simple, upgrade only when needed.
The Relay Race Principle
Complex projects are not marathons with one model. They are relay races. Pass the baton: Opus establishes strategy, Sonnet develops content, Haiku polishes output. Each model runs its best leg. Where Opus still earns its premium: Agent Teams (parallel sub-task coordination unavailable in Sonnet), 128K max output tokens (double Sonnet's ceiling for long-form generation), and first-attempt correctness on ambiguous, multi-step problems.
Three-Question Decision Method
1
How complex is the reasoning required?
Multi-variable analysis, nuanceOpus
Standard business logicSonnet
Clear instructions, simple taskHaiku
2
How many back-and-forth exchanges?
1-5 (Quick analysis)Opus
10-50 (Iterative work)Sonnet
1-3 (Single task)Haiku
3
What is the cost of getting it wrong?
High (Board, investors, strategy)Opus
Medium (Team, clients, quality)Sonnet
Low (Internal, drafts, routine)Haiku
Strategic Work: Use Opus
Strategic Analysis
Complex reasoning, high stakes decisions
2-3 exchanges
Board Q&A Preparation
Political nuance required, anticipate challenges
1-2 per question
Competitive Positioning
Sophisticated market insight needed
3-5 exchanges
Development Work: Use Sonnet
Executive Summary
Strategic synthesis, not simple compression
3-5 exchanges
Board Presentation
Multiple iterations, stakeholder alignment
8-12 exchanges
Business Plan Creation
Comprehensive document, ongoing refinement
10-15 exchanges
Client Proposal Development
Customization, multiple review cycles
6-10 exchanges
Product Documentation
Detailed specifications, technical accuracy
20-30 exchanges
Fast Execution Work: Use Haiku
Meeting Notes Summary
Extract key points, action items
1 exchange
Email Responses
Professional tone, standard templates
1-2 exchanges
Classification and Extraction
Categorize inputs, pull structured data from documents
1-2 exchanges
Data Formatting
Convert formats, clean data, enforce consistency
1-3 exchanges
Quick Q&A and Lookups
Factual answers, definitions, straightforward research
1 exchange
Thinking Executive Summaries are Simple
Using Haiku because "it is just a summary" misses that synthesis requires strategic thinking.
Executive summaries need Sonnet: they require compression plus strategic framing.
Using Opus for Everything
Burning through conversation limits on simple tasks that Haiku could handle.
Save Opus for complex reasoning and high-stakes decisions.
Using Haiku for Strategic Work
Getting shallow analysis when you need deep insight for important decisions.
Switch to Opus when reasoning complexity increases.
Ignoring Exchange Patterns
Starting iterative development projects in Opus and hitting limits quickly.
Use Sonnet for projects requiring 10 or more exchanges.
Not Switching Mid-Project
Staying in one model when the work type changes during a project.
Move between models as task requirements evolve.
The Relay Race Strategy
Opus
→
Framework and Strategy
Sonnet
→
Development and Iteration
Haiku
Polish and Format
Start with Opus to establish the strategic framework, hand off to Sonnet for the heavy lifting of development and iteration, then use Haiku for final formatting and quick edits. Each model plays to its strengths. Note that Opus retains specific advantages unavailable in Sonnet: Agent Teams for parallel task coordination and a 128K output token ceiling (double Sonnet's 64K).
Optimal Usage Distribution
70% of your work should use Sonnet (and the case keeps getting stronger)
Most business work involves iterative development: presentations, proposals, plans, and documents that require multiple rounds of refinement. Sonnet 4.6 now performs within 1-2 points of Opus on most benchmarks while costing significantly less, which means the 70% estimate from November 2025 may actually understate it.
The Context Check-In
Both Opus and Sonnet now support 1M token context windows with automatic compaction, so the old "15-exchange limit" is no longer a hard constraint. But the underlying question is still worth asking periodically: am I using the right model for what this conversation has become?
In Opus? If the work has shifted to iterative refinement, switch to Sonnet.
In Sonnet? If you are hitting the edges of its reasoning, consider an Opus checkpoint.
In Haiku? If you need more depth than you are getting, upgrade.
In Sonnet? If you are hitting the edges of its reasoning, consider an Opus checkpoint.
In Haiku? If you need more depth than you are getting, upgrade.
Match Model to Cognitive Work Type
Choose models based on the type of thinking required, not the perceived importance of the task.
An "important" email might only need Haiku (clear execution).
A "routine" strategy analysis might need Opus (complex reasoning).
Most "critical" presentations need Sonnet (iterative refinement).
A "routine" strategy analysis might need Opus (complex reasoning).
Most "critical" presentations need Sonnet (iterative refinement).
The importance of the output does not determine the model. The cognitive complexity and iteration pattern does. Both Opus and Sonnet now feature adaptive reasoning that automatically scales thinking depth to the prompt, which means the models themselves are getting better at allocating effort. Your job is to put them in the right starting position.
Quick Rule: Start with the simplest model that can handle your task. Upgrade only when needed. As models improve, that threshold keeps moving.
AI Model Quick Reference
Match your task to the right model in 30 seconds. Haiku 4.5 is more capable than most people expect: it rivals earlier Sonnet versions on many tasks.
Exchange counts below reflect a single conversation session. Longer projects span multiple sessions inside a Claude Project — use the markdown handoff technique to carry context across sessions without hitting limits.
Opus Tasks
Strategic Analysis
Complex reasoning, multiple scenarios
5-10 exchanges
Market Research
Deep synthesis, pattern finding
8-15 exchanges
Board Q&A Prep
Anticipate edge cases, defend positions
10-20 exchanges
Complex Problem Solving
Multi-variable analysis, nuanced thinking
10-25 exchanges
Sonnet Tasks
Presentation Development
Iterative refinement, visual design
15-30 exchanges
Client Proposals
Multiple revisions, stakeholder input
20-40 exchanges
Executive Summaries
Polish and precision, multiple drafts
10-20 exchanges
Business Reviews
Comprehensive analysis, ongoing work
25-50 exchanges
Haiku Tasks
Email Drafting
Clear templates, quick turnaround
1-2 exchanges
Classification and Extraction
Categorize inputs, pull structured data
1-2 exchanges
Document Formatting
Structured output, consistent style
2-3 exchanges
Quick Q&A and Lookups
Factual answers, definitions, research
1 exchange
Quick Edits
Fast fixes, simple changes
1 exchange