Three Models, Three Purposes
Claude isn't one model - it's a family of three, each designed for different jobs. Choosing the right one is the difference between a $5 API bill and a $500 one.
⚡ Haiku (Speed)
The fastest, cheapest model. Instant responses for simple tasks. Think of it as your "quick assistant" - great for classification, extraction, and simple Q&A.
🎶 Sonnet (Balance)
The sweet spot. Smart enough for complex tasks, fast enough for real-time use. This is what most production systems should use. Best price-to-performance ratio.
🎧 Opus (Power)
The most capable model. Deep reasoning, complex analysis, nuanced writing. Use when quality matters more than speed or cost. The "senior engineer" of the family.
Pricing Breakdown (Per Million Tokens)
| Model | Input | Output | Speed | Intelligence |
|---|---|---|---|---|
| Haiku 4.5 | $1.00 | $5.00 | ⚡⚡⚡⚡⚡ | ⭐⭐⭐ |
| Sonnet 4.6 | $3.00 | $15.00 | ⚡⚡⚡⚡ | ⭐⭐⭐⭐ |
| Opus 4.6 | $5.00 | $25.00 | ⚡⚡ | ⭐⭐⭐⭐⭐ |
Haiku is 5x cheaper than Opus and 3x cheaper than Sonnet. Sonnet handles 90% of use cases. Opus 4.6 is surprisingly affordable now - only 1.67x the cost of Sonnet for input.
The "Brain + Muscles" Pattern
The smartest teams don't pick one model - they use multiple models together. This is the most important architecture pattern in production AI:
🧠 Brain (Opus/Sonnet)
Makes decisions, plans strategy, handles complex reasoning. Called rarely - maybe once per workflow. Expensive but worth it for the hard stuff.
💪 Muscles (Haiku)
Executes the plan. Runs classification, extraction, formatting on hundreds of items. Called frequently - speed and cost matter here.
Email triage system: Haiku classifies 1,000 incoming emails (fast, cheap). Only the 50 flagged as "complex" get sent to Sonnet for detailed analysis. Result: 95% cost reduction vs sending everything to Sonnet.
Decision Framework: Which Model When
| Task | Model | Why |
|---|---|---|
| Classify text (spam/not spam) | Haiku | Simple binary task, speed matters |
| Extract data from invoices | Haiku | Structured extraction, high volume |
| Write marketing copy | Sonnet | Needs creativity + brand voice |
| Code review | Sonnet | Needs understanding, not just pattern matching |
| Build a full feature from spec | Opus | Complex multi-file reasoning |
| Legal document analysis | Opus | Nuance matters, mistakes are expensive |
| Real-time chatbot | Sonnet | Balance of quality and response time |
| Monitoring/alerts | Haiku | Running 24/7, cost adds up fast |
Common Mistakes
"It's the best model, so I should always use it." Wrong. You're burning 60x more money than Haiku for tasks that don't need it. Match the model to the task complexity.
Haiku is fast but it cuts corners on nuanced analysis. If your task requires understanding context across a long document or making judgment calls, step up to Sonnet.
Always test your prompt on all three models. You might be surprised - Sonnet often matches Opus on tasks you assumed needed the bigger model. Test with real data, not toy examples.
Key Takeaways
- Haiku = speed + cost. Use for classification, extraction, simple tasks
- Sonnet = balance. Default choice for 90% of production use cases
- Opus = power. Complex reasoning, high-stakes analysis
- Brain + Muscles pattern: Opus decides, Haiku executes = best of both
- Always test across models - you'll often find Sonnet is enough
- Cost difference is 60x between Haiku and Opus - model choice IS cost optimization