Model families and choosing one
Claude models come in families, each sitting at a different point in the quality, speed, price triangle. Names and prices change, so rather than memorising figures it is more useful to internalise the selection principle:
- The Opus family — the strongest reasoning. Complex refactors, architectural decisions, long agentic tasks.
- The Sonnet family — the balanced choice. Most day-to-day coding.
- The Haiku family — fastest and cheapest. Simple mechanical tasks, subagents, bulk operations.
Pricing is counted separately for input and output tokens, and output is always the more expensive of the two.
| Task | Fitting choice | Reason |
|---|---|---|
| XML → Compose migration, multi-file refactor | The strongest model | Requires long context and multi-step planning |
| Day-to-day feature work, bug fixes | The balanced model | Quality is sufficient, cost is lower |
| Translating string resources, formatting fixes | The fastest model | Mechanical work — no judgement needed |
| Code review, security analysis | The strongest model | The cost of a missed bug exceeds the cost of tokens |
Dropping to a cheaper model to save cost is a team/business decision, not a silent one. A practical rule: when a mistake is expensive (production, security, migration) use the strong model; when a mistake is cheap and immediately visible (a small change that will fail to compile) a cheaper model is fine.
In Claude Code you can switch models inside a session with /model. Beyond that:
- effort controls how much "thinking" the model spends on a task. Low effort is cheap and fast; high effort is for hard tasks.
- Subagents can be given their own cheaper model — the main agent plans while subagents do mechanical work such as searching.
📚 Sources and documentation
- Claude models overviewofficialplatform.claude.com
Current model names, context limits and capabilities.
- Model and effort configurationofficialcode.claude.com