With Opus 5.5 (22 September 2026), Sonnet 5.5 (28 September) and Haiku 5.5 (7 October), Anthropic now has a full 5.5 family. The question for a business is not which is best. It is which is the cheapest one that does each job well enough.
The listed prices
- Haiku 5.5: $0.10 input and $0.50 output per million tokens (prompts up to 100,000 tokens).
- Sonnet 5.5: $2 input and $10 output per million tokens (prompts up to 100,000 tokens).
- Opus 5.5: $4 input and $20 output per million tokens.
Prices change, so check Anthropic's pricing page before you budget.
A plain way to choose
| Job | Start with |
|---|---|
| Sort, label, extract, route, short FAQ answers | Haiku 5.5 |
| Everyday drafting, analysis, coding help, most agent work | Sonnet 5.5 |
| Hard multi-step reasoning, complex coding, computer use, high-stakes work | Opus 5.5 |
The routing pattern
Most good business agents do not pick one model. They route:
- A small model reads the request and decides what kind of task it is.
- Routine tasks stay with the small model.
- Harder tasks, or tasks where the small model is unsure, go up one size.
- Anything risky, such as sending money or messaging a customer, gets a check by a person.
This keeps the average cost close to the small model while keeping quality close to the big one where it matters.
Test, do not guess
Take 30 to 50 real examples of your task. Run them through two sizes with the same instructions and score the results. If the cheaper model is close, use it. This takes an afternoon and often saves a large share of the model bill.
One more tip: keep the model swappable
Models are replaced every few months. Build your agent so the model name is a setting, not baked into the code. When a better or cheaper model appears, you can test and switch in an afternoon.
Want an AI agent built around your own business, with the right model for each job? Talk to Power Digital and we will scope it with you.
