Claude Haiku 5.5 API pricing
Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens at the standard rate. Here are its cached and batch prices, and what it costs for your usage.
| Token type | Standard | Batch (50% off) |
|---|---|---|
| Input | $0.10 | $0.05 |
| Cached input | $0.01 | $0.005 |
| Output | $0.50 | $0.25 |
Claude Haiku 5.5: Prompts over 100,000 tokens are billed at a higher rate.
- Input $3.00
- Cached input $0.3
- Output $4.50
| Workload | Standard | Batch | |
|---|---|---|---|
Support chatbot 2,000 in / 300 out tokens,
1,000 requests a day, 50% cached | $7.80 | $3.90 | 2nd cheapest of 17 |
Document summaries 10,000 in / 800 out tokens,
200 requests a day | $8.40 | $4.20 | 2nd cheapest of 17 |
Tagging / classifying 400 in / 20 out tokens,
20,000 requests a day | $30.00 | $15.00 | 2nd cheapest of 17 |
Compare Claude Haiku 5.5 with
Other Anthropic models
FAQ
Frequently asked questions
How much does Claude Haiku 5.5 cost per million tokens?
Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens at the standard rate. Prices are in USD, before tax, from Claude API pricing, last checked on Oct 13, 2026. Prompts over 100,000 tokens are billed at a higher rate.
Does Claude Haiku 5.5 have a Batch API discount?
Yes. Through the Batch API, Claude Haiku 5.5 costs $0.05 per million input tokens and $0.25 per million output tokens, 50% less than the standard rate. Batch requests are processed asynchronously rather than straight away.
How much does prompt caching save on Claude Haiku 5.5?
Input read from the prompt cache costs $0.01 per million tokens instead of $0.10, 90% less. Writing to or storing the cache can cost extra; those charges aren’t included here.
Is Claude Haiku 5.5 one of the cheaper API models?
For a support chatbot sending 1,000 requests a day (2,000 input and 300 output tokens each, 50% of the input cached), Claude Haiku 5.5 is the 2nd cheapest of the 17 OpenAI, Anthropic and Google models we track, at $7.80 a month. Price alone doesn’t tell you how well a model will handle your task, so test the candidates on your own prompts.