Opus vs Sonnet for Claude Code - Choosing the Right Model for Each Command
Opus vs Sonnet - Picking the Right Model for Each Task
Not every Claude Code command needs the most powerful model. Using Opus for everything is like driving a truck to the grocery store - it works, but you're burning fuel you don't need.
Sonnet Is Fine for Init
The /init command scans your project structure and generates a CLAUDE.md file. This is mostly pattern recognition and file listing - exactly what Sonnet handles well. The output is a configuration file, not complex code. Save your Opus tokens for the work that follows.
Other tasks where Sonnet performs comparably:
- File exploration - reading and summarizing existing code
- Simple refactors - renaming variables, extracting functions
- Documentation generation - writing comments and READMEs
- Test scaffolding - generating boilerplate test files
Opus Shines for Implementation
Where Opus earns its cost is in multi-step implementation tasks that require holding complex context and making architectural decisions:
- Cross-file refactoring that touches 10+ files with interdependencies
- Bug diagnosis in unfamiliar codebases where reasoning chains matter
- System design decisions that need to consider tradeoffs
- Complex feature implementation spanning multiple modules
The difference shows up in subtle ways. Opus is more likely to ask clarifying questions before diving in, consider edge cases unprompted, and maintain consistency across a long implementation session.
A Practical Split
A reasonable default: use Sonnet for everything except active implementation and debugging. That covers planning, exploration, documentation, and simple changes with the cheaper model. Switch to Opus when you're about to write or modify significant logic.
Some teams automate this with model routing - the orchestrator session runs Sonnet while implementation sessions run Opus. The orchestrator doesn't need deep reasoning. It just needs to break tasks apart and check completion.
The Cost Math
Opus costs roughly 5x more than Sonnet per token. If 60% of your Claude Code usage is exploration and planning, switching those tasks to Sonnet cuts your bill by nearly half with no quality loss on the work that matters.
Fazm is an open source macOS AI agent. Open source on GitHub.