Claude Code lets you switch between models. The choice affects speed, cost, and quality. Here's how I think about it.
The models available
As of early 2026: Claude Haiku (fast, cheap), Claude Sonnet (balanced), Claude Opus (most capable, slowest, most expensive). The names change as new versions release — the tradeoffs stay the same.
Fast/cheap model (Haiku) — use for
- Code search and exploration ("find all usages of this function")
- Simple transformations (rename, reformat, add types)
- Generating boilerplate with a clear template
- Writing tests for simple functions
- Quick questions about what code does
For these tasks, the fastest model is accurate and you don't pay for capability you don't need.
Balanced model (Sonnet) — the default
- New feature implementation
- Bug fixing
- Refactoring
- Writing tests for complex code
- Most day-to-day coding tasks
Sonnet is the default for a reason. Good capability, reasonable speed, reasonable cost.
Most capable model (Opus) — use for
- Complex architectural decisions
- Debugging hard problems where you've tried and failed
- Security review of critical code
- Tasks where you've tried Sonnet and gotten wrong results
- Understanding a complex unfamiliar codebase
Opus is slower and more expensive. The quality difference is real but not always necessary. Use it when Sonnet is getting it wrong.
The practical approach
Default to Sonnet. Switch to Haiku for exploratory work (file reading, searching, summarizing). Switch to Opus when Sonnet produces bad results on the same task twice.
Don't default to Opus. The speed difference is significant in a work session and the quality improvement isn't always proportional to the cost.
Top comments (0)