The short answer
- Sonnet — your everyday default. Handles most coding well at a fraction of Opus's cost.
- Opus — the hard stuff: big refactors, gnarly debugging, architecture. Worth its premium only when the task earns it.
- Haiku — fast and cheap for simple, mechanical work where top-tier reasoning is overkill.
- The trap: running Opus on routine tasks. Same job, several times the price — the most common quiet overspend.
The three models at a glance
| Model | Strength | Relative cost | Use it for |
|---|---|---|---|
| Opus | Deepest reasoning | highest | Big refactors, hard bugs, architecture |
| Sonnet | Balanced — the default | mid | Everyday coding, most tasks |
| Haiku | Fast & cheap | lowest | Simple, mechanical, high-volume work |
The tiers span several times in price from bottom to top. Combined with output costing several times more than input, model choice alone can swing a task's cost dramatically. Confirm current per-token rates at claude.com/pricing — they change.
When to reach for Opus
Opus earns its premium when the problem is genuinely hard and a wrong answer is expensive:
- Large, cross-cutting refactors where the model has to hold a lot in its head.
- Subtle debugging where cheaper models keep guessing wrong.
- Architecture and design decisions you'll live with for a while.
- Anything where one good answer beats three cheap-but-wrong attempts.
That last point is the real test: if a cheaper model would need several retries to land it, the "expensive" model can be cheaper overall.
When Sonnet (or Haiku) is the smart pick
Most coding isn't the hard case. Writing a component, wiring an endpoint, fixing a clear bug, editing docs — Sonnet handles these well, and Haiku is great for the truly mechanical, high-volume stuff. Reaching for Opus here is like taking a taxi to the mailbox: it works, you just paid several times over for no benefit.
/model to change models mid-session. Most overspend
isn't a big mistake; it's Opus quietly doing Sonnet's job all day.
Model choice is one of four cost levers
It's the biggest, but not the only one. The full set: model, context size (re-sent every turn), thinking/effort, and tool loops. We break all four down in how much Claude Code costs and how to reduce it.
See whether your model mix is wasteful
The problem with model choice is that the cost of a bad default is invisible in the moment — you only feel it at the end of the month. Dinoradar reads your local session history and breaks spend down so you can see where the money goes, making it obvious when an expensive model has been doing cheap work.
Frequently asked
Should I use Opus or Sonnet in Claude Code?+
Use Sonnet as your everyday default — it handles most coding well at much lower cost. Reach for Opus on genuinely hard problems: big refactors, tricky debugging, architecture. Since Opus costs several times more per token, running it on routine work is the most common way people overspend.
What is the default model in Claude Code?+
Claude Code uses a capable default suited to general coding and lets you switch per session with /model. What matters isn't today's default but that model choice is yours — and it's one of the biggest levers on both quality and cost.
How much cheaper is Sonnet than Opus?+
Per token, the top model costs several times more than the mid tier, which costs several times more than the smallest. The same task can swing several times in price on model choice alone. Confirm current per-token rates at claude.com/pricing, since they change.
How do I switch models in Claude Code?+
Run /model inside a session. A good habit: start on Sonnet and switch up to Opus only when a task proves hard, then switch back. Dinoradar breaks your spend down so you can spot when expensive models are doing cheap work.
Find out where your model choice costs you.
Dinoradar reads your history locally and shows spend per project and over time — so a lazy Opus default has nowhere to hide. Free forever · Pro $29 once.