Which Claude Model Should I Use?

The Three Models

Claude Code runs on three AI models. Think of them as three colleagues with different strengths:

All three support 1 million token context windows and extended thinking — meaning they can reason through complex problems step by step before answering.

When to Use Opus 4.6

Use Opus when the stakes are high or the problem is complex.

• Designing system architecture or making foundational decisions
• Refactoring that touches 10+ files
• Debugging cross-file issues where your first approach failed
• Writing PRDs, competitive analysis, or strategy documents
• Complex multi-step reasoning that requires getting it right the first time
• Security reviews and code audits

Opus is the "senior engineer" — slower, more expensive, but the quality ceiling is higher. When you need it, you really need it.

Pricing: $5 input / $25 output per million tokens.

When to Use Sonnet 4.6

Use Sonnet for almost everything else. This is your daily driver.

• Single-file edits and bug fixes
• Writing tests, documentation, and reviews
• CSS tweaks and UI changes
• Git operations and daily workflow
• Code generation for straightforward features
• Answering questions about your codebase
• Content writing when the strategy is already decided

Sonnet is faster and cheaper, and for 80% of tasks the quality difference from Opus is negligible.

Pricing: $3 input / $15 output per million tokens.

When to Use Haiku 4.5

Use Haiku for speed-critical or high-volume tasks.

• Quick code formatting
• Simple lookups ("what does this function do?")
• Running through checklists
• Basic text transformations
• Syntax checks

Haiku is the "intern with a fast typing speed" — great for volume work where you don't need deep reasoning.

Pricing: The cheapest option. Good for tasks where speed matters more than depth.

How Claude Code Picks For You

If you're on Claude Pro or Max, you generally don't need to pick manually. Claude Code routes based on your plan:

• Pro ($20/month): Defaults to Sonnet. You get generous daily usage. Good for most people.
• Max ($100/month): Defaults to Sonnet with access to Opus. 20x the usage of Pro. For power users.
• API (pay per use): Full control over which model to use. You pick per request.

Switching Models Mid-Session

You can switch models at any time:

# In the terminal
/model opus     # Switch to Opus for a complex task
/model sonnet   # Switch back for regular work
/model haiku    # Switch to Haiku for quick tasks

Which Plan Should I Get?

Start with Pro. Most people never need Max. You'll know if you need it because you'll hit the daily limit and feel frustrated.

Pro Tips

1. Start every task on Sonnet. Only switch to Opus if Sonnet's first attempt doesn't cut it.
2. Use Haiku for sub-agents. If you're running code review or test agents in the background, Haiku is fast and cheap.
3. Watch your context window. A 1M token context sounds huge, but if you're loading entire codebases, costs add up. Use /compact to trim.
4. Extended thinking is free in terms of input. The model "thinks" before answering but you only pay for the output tokens.

The Decision Flowchart

Is this task complex (architecture, 10+ files, strategy)?
├── Yes → Use Opus
└── No
    Is this a quick formatting/lookup task?
    ├── Yes → Use Haiku
    └── No → Use Sonnet (default)

Keep it simple. Sonnet for 80% of work. Opus for the hard stuff. Haiku for the fast stuff.