Claude Haiku 4.5 API pricing
Last verified
Claude Haiku 4.5 costs $1.00 per million input tokens and $5.00 per million output tokens on Anthropic's first-party API.
Rates
| Token type | $ per million |
|---|---|
| Input | $1.00 |
| Output | $5.00 |
| Cached input | $0.10 |
| Cache write (5 min) | $1.25 |
| Cache write (1 hour) | $2.00 |
| Batch input | $0.50 |
| Batch output | $2.50 |
Output costs 5.0× input. That ratio is what decides whether a workload is expensive: reading a lot is cheap, generating a lot is not. Weight your estimates accordingly.
What it costs to run something
Monthly cost at Claude Haiku 4.5's rates, assuming no caching. Token figures are stated so you can check the arithmetic.
| Workload | Monthly cost |
|---|---|
| Support chatbot 1,000 conversations/month · 2,000 in + 500 out each | $4.50 |
| RAG search 10,000 queries/month · 8,000 in + 400 out each | $100 |
| Bulk extraction 10,000 documents · 3,000 in + 300 out each | $45.00 |
| Coding agent 100 runs/month · 400,000 in + 20,000 out each | $50.00 |
With prompt caching on a stable prefix, the input side of each of these falls toward $0.10 per million — frequently a majority of the bill.
Caching economics for Claude Haiku 4.5
Cached input is billed at $0.10 per million tokens — 0.10× the base input rate. For a workload that resends the same prefix, that is the single largest lever available.
| Operation | $ / Mtok | vs base input | Breaks even after |
|---|---|---|---|
| 5-minute cache write | $1.25 | 1.25× | 1 read |
| 1-hour cache write | $2.00 | 2.00× | 2 reads |
| Cache read | $0.10 | 0.10× | — |
A write costs more than fresh input, so caching only pays once the content is read back. In an interactive session that happens immediately; in a one-shot request it does not.
Batch pricing
Asynchronous batch processing is billed at $0.50 in / $2.50 out per million tokens — a 50% discount. Right for anything a user is not waiting on: bulk classification, evaluation runs, backfills, embedding a corpus.
Where it sits, and what's cheaper
Among Anthropic's current models, Claude Haiku 4.5 is #1 of 10 by input price (1 = cheapest).
Cheaper models that may do the job — worth testing before you assume you need this tier:
| Model | Input | Output | Saving on input |
|---|---|---|---|
| GPT-5.4 mini | $0.75 | $4.50 | 25% |
| Gemini 3 Flash Preview | $0.50 | $3.00 | 50% |
| GPT-4.1 mini | $0.40 | $1.60 | 60% |
The honest caveat: a cheaper model is only cheaper if it still does the task. Test on your actual workload rather than assuming, and measure quality before you migrate.
Tracking what you actually spend
These are list prices. What you are billed also reflects credits, discounts, caching and batch usage — which is why the number worth watching comes from Anthropic's billing API, not a calculator.
How to track spend across providers · Why billing data is daily
Source
Rates verified 2026-08-05 against Anthropic's pricing documentation. Prices change — if this page is more than a month old, check the source.