The costliest habit is sending every task to the flagship. Choosing a model on purpose is easy, and the rule for it is short.
The rule in one line
The flagship is for jobs that need many relationships held at once. Everything else is within reach of a smaller model.
If a task can be put as "take this and turn it into that", you do not need the flagship. If it cannot be solved without understanding how the whole system fits together, you do.
Where Opus earns its price
- reading an unfamiliar repository to understand its architecture
- refactoring that touches several files coherently
- hunting races and intermittent bugs
- design work that weighs trade-offs
- long agentic runs where the model picks its own next step
Where Opus means overpaying
- parsing logs and traces
- rewriting and proofreading text
- boilerplate code generated from a pattern
- classifying inbound messages
- format conversion, parsing, data normalisation
- simple questions about documentation
On jobs like these a smaller model does just as well, and the bill comes out several times lower.
The two-model trick
A scheme that works: the flagship thinks, a smaller model executes.
Opus first breaks the task down and writes a plan. Sonnet or gpt-5.6-luna then carries out the mechanical part against that plan. You pay for the expensive reasoning once, not at every step.
Haiku is in the catalog: claude-haiku-5.5 and claude-haiku-4.5, with prices on the models page. They are available in the Premium channel and absent from the Cheap channel.
The trick shows best on batch work: work out the approach once, then apply it to a thousand items cheaply.
About input and output
One more detail people often miss: on every model, output costs several times more than input.
So a long context in the prompt is almost always cheaper than a long answer. Do not be afraid of giving context. Be afraid of asking for a wordy reply.
Check it against your own numbers
General reasoning translates poorly into money. The site has a calculator: pick a model, enter volumes, get the sum. Run the same scenario across tiers. The gap usually persuades better than any argument.
In short
The flagship ties context together, a smaller model handles transformations. Thinking is expensive and doing is cheap, so split those two jobs between different models.