- Sonnet for everyday work. It’s fast, and you get more of it.
- Opus when being wrong is expensive: plans, reviews, big decisions.
- The best setup uses both: the big one plans and checks, the smaller one does the work.
The brain and the hands
Opus
The brain
- Slower, and costs more
- Plans work with many parts
- Reviews work before it goes out
Sonnet
The hands
- Fast, and cheaper
- Drafts, summaries, rewording
- The routine work you repeat
Which one for which job?
| The job | Use |
|---|---|
| Drafting an email or a post | Sonnet |
| Summarizing a document | Sonnet |
| Planning a launch with many parts | Opus |
| Checking work before a client sees it | Opus |
| A decision with real money on it | Opus |
| Simple jobs, lots of them | Haiku, the smallest |
In July I was building a planning tool with AI, and I made a rule.
Use the big model to plan and test. The boss. Use the smaller one as the worker. And check all its work.
My assistant runs the same way. Ordinary messages go to the everyday model. If I write “think:”, it switches up.
The most expensive model has a daily cap.
Opus vs Sonnet at a glance
| Opus | Sonnet | |
|---|---|---|
| Speed | Slower | Faster |
| Cost and limits | Costs more. Limits arrive sooner. | Cheaper. Goes further. |
| Best at | Planning, judging, hard problems | Everyday drafting and routine work |
| The role I give it | The boss | The worker |
Three mistakes I see people make
- Using the big model for everything.You hit your limit by lunchtime, and simple answers aren’t any better.
- Letting the smaller model check its own work.The check should come from the stronger model, or from a plain rule.
- Switching models before fixing the request.A thin answer usually means a thin request.
How to choose, step by step
Talk it through with me. Tell me what your work looks like and we’ll find where AI helps first.
Quick answers
What is the difference between Opus and Sonnet?
Size. Opus is the larger Claude model: slower, more expensive, better at hard reasoning. Sonnet is the mid-sized one: faster, cheaper, and right for most work.
When should I use Opus?
When being wrong is expensive: a plan with many parts, a review before something goes out, a decision with money on it.
It thinks further ahead and catches more of its own mistakes.
When is Sonnet enough?
Most of the time. Drafts, summaries, rewording, questions about a file.
If an answer comes back thin, that’s your signal to switch up. Not before.
Is the bigger model always better?
No. It’s slower, and you run out sooner.
My rule: a cheaper model can do the work, but it never gets the last word on what reaches me.
Which Claude model is best for writing?
Sonnet for most writing. Switch to Opus for a long piece with a lot to hold together, or for the final review.
What is Claude Haiku?
The smallest and fastest Claude model. It suits simple, repeated jobs where speed matters more than depth.
The names and version numbers change a few times a year, and there is a tier above Opus as I write this. The pattern stays: small, medium, large.
Next: the best Claude model for each job, Fable 5.2 vs Opus 5.5, and Claude vs ChatGPT.
AI Foundations starts here: give AI a clear picture of your business in a form you can reuse, so you stop explaining from zero.
Background: Anthropic: Claude models