I use Claude Haiku 4.5 for small refactors, code review, and debugging. Those jobs benefit more from quick responses and lower cost than from the deeper reasoning I would pay Sonnet for.
The Numbers
| Model | Input (1M tokens) | Output (1M tokens) |
|---|---|---|
| Haiku 4.5 | $1 | $5 |
| Sonnet 4.5 | $3 | $15 |
Anthropic launched Haiku 4.5 at $1 per million input tokens and $5 per million output tokens. Sonnet 4.5 was $3 and $15. Haiku cost one-third as much on both sides, which mattered when a workflow made many calls.
Where I Use It
For these bounded tasks, the faster response makes the tool feel more immediate. Sonnet earns its price when I need complex reasoning or nuanced writing.
When Cost Decides
Haiku made repeated small calls easier to justify. I did not use it when the task needed architectural judgment or a long debugging chain; that is where paying for Sonnet saved retries. The cheaper model earned a place by staying inside a narrow job.
One quick signal
Did this earn your time?
Thanks. That gives me something concrete to check.
About the author
KahWee Teng
I’m a staff software engineer in San Francisco. I write from systems I’ve built and maintained: frontend migrations, AI-assisted workflows, and hardware running at home.
