A worked estimate you can check line by line. Change the numbers to your own in the cost calculator.
What we assume
2,000 tokens of instructions and examples (identical on every call), about 600 tokens per item (an email, a review, a form), and a 100-token structured answer.
Cost per item
| Model | No caching | With caching | Caching + Batch API | Per 1,000 items, with caching |
|---|---|---|---|---|
| Claude Fable 5.2 | $0.025 | $0.0092 | $0.0046 | $9.20 |
| Claude Opus 5.5 | $0.012 | $0.0048 | $0.0024 | $4.80 |
| Claude Sonnet 5.5 | $0.0062 | $0.0024 | $0.0012 | $2.40 |
| Claude Haiku 5.5 | $0.00031 | $0.00013 | $0.000065 | $0.130 |
Cost per month
| Model | 10,000 items/month | 1,000,000 items/month |
|---|---|---|
| Fable 5.2 | $46.00 | $4,600 |
| Opus 5.5 | $24.00 | $2,400 |
| Sonnet 5.5 | $12.00 | $1,200 |
| Haiku 5.5 | $0.650 | $65.00 |
Where the money goes
- This is the job small models are made for. Test the smallest model first with a few hundred labelled examples.
- The instructions and examples repeat on every item, so caching and the Batch API stack well here.
- Use structured outputs so every answer parses; a retry for a broken answer doubles that item's cost.
To check what a plan price leaves you after these costs, use the AI SaaS margin calculator. To see how much caching saves on your own prompt sizes, use the prompt caching savings calculator.
Questions
How much does one item cost with the Claude API?
With prompt caching, from about $0.00013 on Claude Haiku 5.5 to about $0.0092 on Claude Fable 5.2, for a item of 2,000 reused + 600 new input tokens, 100 output tokens. Without caching it is $0.00031 to $0.025.
How much does it cost per month?
10,000 items/month: about $0.650 on Haiku 5.5 up to $46.00 on Fable 5.2. 1,000,000 items/month: about $65.00 on Haiku 5.5 up to $4,600 on Fable 5.2, using the Batch API, with caching.
How can I make it cheaper?
This is the job small models are made for. Test the smallest model first with a few hundred labelled examples.