Claude Haiku 4.5 API pricing is $1 per million input tokens and $5 per million output tokens (USD). That is token-based developer API pricing—not the price of a Claude app subscription. Anthropic lists these rates and a context window of up to 200,000 tokens in its Claude Haiku 4.5 model documentation. (Claude Haiku 4.5 model documentation)
Model-status note (November 9, 2026): Anthropic’s model page labels Haiku 4.5 Active (legacy) and says it is still available, while recommending that users consider migrating to Haiku 5.5. The page’s retirement field says “not sooner than October 15, 2026”; it does not give a specific scheduled retirement date. Check the current documentation and your chosen API channel before committing a production budget. (Claude Haiku 4.5 model documentation; Anthropic’s model lifecycle definitions)
Estimate your Haiku 4.5 API cost
Use this formula for standard, uncached token usage:
(input tokens ÷ 1,000,000 × $1)
+ (output tokens ÷ 1,000,000 × $5)
= estimated API token cost in USD
Example: For 10 million input tokens and 2 million output tokens in a month:
- Input: 10 × $1 = $10
- Output: 2 × $5 = $10
- Estimated total: $20
This is a rate-based estimate using Anthropic’s listed prices, not a guaranteed invoice. It does not include cache or batch adjustments, token usage or separate charges associated with optional tools, taxes, or provider-specific charges. Whether any such items apply depends on the features and billing channel you use. A third-party pricing page gives the same 10M-input/2M-output example as $20; the arithmetic above follows directly from Anthropic’s published rates. (Anthropic pricing documentation; third-party example)
For your own estimate, substitute the total input and output tokens you expect to send during the billing period. Keep the two token totals separate because output tokens cost five times as much per token as standard input tokens at the listed rates.
Published prompt-caching rates
Anthropic lists these Haiku 4.5 cache prices:
| Token category | Price per million tokens |
|---|---|
| Standard input | $1.00 |
| Standard output | $5.00 |
| 5-minute cache write | $1.25 |
| 1-hour cache write | $2.00 |
| Cache hit or refresh | $0.10 |
A cache write and a cache hit are distinct billed categories; don’t price every input token at the cached rate by default. Whether caching changes your estimate depends on whether you reuse eligible context and which cache operation applies. (Claude Haiku 4.5 model documentation)
Batch pricing for asynchronous requests
Anthropic’s Haiku 4.5 documentation says Batch API requests receive a 50% discount on input and output. Applying that discount to the listed standard rates gives $0.50 per million input tokens and $2.50 per million output tokens. Batch requests are processed asynchronously, rather than returned as immediate responses. Anthropic says most batches finish in less than one hour; that is a typical duration, not a guaranteed completion time. Use batch only when your workflow can tolerate asynchronous processing, and confirm current terms before budgeting. (Haiku 4.5 model documentation; Batch processing documentation)
Should you compare Haiku 4.5 with Haiku 5.5?
Yes, if you are choosing a model for a new production feature. Anthropic lists Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens; prompts over 100K have a higher rate card. Those are Haiku 5.5 rates, not a guarantee of savings on your particular workload. (Haiku 5.5 model documentation)
Anthropic originally positioned Haiku 4.5 for low-latency tasks such as chat assistants, customer service, and pair programming. That context does not establish that it will meet your application’s quality or latency requirements. Test models on representative prompts and compare cost per successfully completed task—not just list price per token. (Anthropic’s Haiku 4.5 announcement)
Bottom line
For standard Claude Haiku 4.5 API usage, estimate from $1 per million input tokens and $5 per million output tokens: 10M input plus 2M output tokens works out to $20 before applicable modifiers. Anthropic also documents cache prices and a 50% Batch API discount for input and output. Batch processing is asynchronous, and most batches finish in less than an hour, but that is not a completion guarantee. Anthropic currently lists Haiku 4.5 as Active (legacy) and still available; check its live lifecycle documentation and any provider-specific terms before deployment.
Sources
Claude Haiku 4.5 — Claude Platform Docs
Model deprecations — Claude Platform Docs
Pricing — Claude Platform Docs
Batch processing — Claude Platform Docs
Claude Haiku 5.5 — Claude Platform Docs