Claude API for business: cost and use in Australia
A Claude plan is for people. The API is for software. Here is what the API costs per job in Australian dollars and where your data is processed.
Short answer: a Claude plan is for people who chat with Claude. The API is for software that calls Claude on its own, such as a script that reads every new HubSpot call note. API use is billed per token, and most back-office jobs cost a few Australian cents or less each.
Below is how the two differ, what the API is used for, what a job costs in Australian dollars, and what we could and could not confirm about where your data is processed. We read Anthropic’s pricing and data residency documentation on 31 October 2026.
Plan or API?
A plan (Pro, Max, Team or Enterprise) gives a person a seat in Claude. They type, Claude answers, and the seat price covers it. The API has no seat and no chat window. Your own software sends Claude a request and gets text back, and you pay for the amount of text that passes through.
Most businesses need plans first. Reach for the API when the same task repeats and a person should not be the one triggering it: summarising every call, sorting every inbound email, checking every invoice. Plan prices are in our Claude pricing guide. This post covers the API side.
What the API gets used for
- HubSpot record clean-up. Read a call transcript, write a three-line summary, and fill the next-step and objection fields on the deal.
- Triage. Read each inbound enquiry, label it (sales, support, supplier, spam) and route it to the right owner.
- Document reading. Pull supplier, amount, due date and line items out of invoices and contracts, ready for a person to approve in the accounts system.
- Drafting. Write a first-pass quote, renewal email or board summary from your own data, left in drafts for someone to review.
- Overnight jobs. Re-check a few thousand contact records for duplicates or missing fields while nobody is working.
Each of these is a small, bounded task. That is why they are cheap.
How token pricing works
The API charges for tokens, which are small chunks of text. A token is roughly three quarters of a word. You pay for input tokens (what you send, including your instructions and the document) and output tokens (what Claude writes back). Output costs five times as much as input per token. Prices are per million tokens, in US dollars; we convert at A$1.42 per US$1.
| Model | Input, US$ | Output, US$ | Input, about A$ | Output, about A$ |
|---|---|---|---|---|
| Haiku 5.5 | $0.10 | $0.50 | $0.14 | $0.71 |
| Sonnet 5.5 | $2 | $10 | $2.84 | $14.20 |
| Opus 5.5 | $4 | $20 | $5.68 | $28.40 |
| Fable 5.1 | $10 | $50 | $14.20 | $71 |
Haiku is the small, fast model for simple sorting. Sonnet handles most business writing and extraction. Opus and Fable are for harder reasoning, and we reach for them only when a cheaper model gets the job wrong. One detail: Haiku 5.5 costs more per token on prompts over 100,000 tokens, which a normal back-office job never reaches.
What a job costs in Australian dollars
These are worked from the prices above, before GST. Your own numbers will differ with the length of your documents.
| Job | Model and size | Cost per job | Cost for the volume |
|---|---|---|---|
| Label an inbound email | Haiku 5.5, 1,000 tokens in, 200 out | About A$0.0003 | 10,000 emails: about A$2.84 |
| Summarise a call into HubSpot | Sonnet 5.5, 3,000 in, 500 out | About A$0.016 | 1,000 calls: about A$16 |
| Read a 20-page contract | Opus 5.5, 30,000 in, 2,000 out | About A$0.23 | 200 contracts: about A$45 |
The pattern is that a full month of a task like this costs less than one Team seat. The expensive part of an API build is the design and testing, which is why the next section matters.
Batch and caching
The Batch API takes jobs that can wait and charges half price on both input and output. Anthropic runs them asynchronously, so use it for overnight work, not for anything a person is waiting on. The 1,000 call summaries above would drop from about A$16 to about A$8.
Prompt caching helps when you send the same long text with every request, such as a 20,000-token policy document that tells Claude how to judge each enquiry. Anthropic stores that text for a short time, and reading it back costs a fraction of normal input. On Sonnet 5.5 a cache read is US$0.10 per million tokens against US$2 for fresh input. Writing to the cache costs 1.25x the base price for the five-minute option, so it pays off after the first reuse.
Where your data is processed
This is the part we checked most carefully. On the Anthropic API you can set inference_geo to global (the default, where Anthropic may run your request in any available region) or us (US infrastructure only). US-only costs 1.1x on every token category. The documentation says those two are the only inference options, and that US is the only workspace storage region.
We could not find an Australian processing option on Anthropic’s own API. Anthropic’s documentation also lists Amazon Bedrock and Google Cloud as other ways to reach Claude, each with its own regional endpoints and pricing. We have not confirmed whether any of those offer processing inside Australia for the current models, so check that before you promise a client or regulator anything. Zero data retention is available for eligible customers on some models, and the same documentation lists HIPAA readiness; neither is automatic.
If data location matters to your business, decide it before the build. It affects which platform we use.
Who should build it
A developer can set up a simple API call in an afternoon. The harder work is deciding what Claude should see, what it may write, what a person approves, and how you find out when it gets something wrong. Without that, a cheap script can overwrite good HubSpot data and nobody notices.
We’re Certified in the Claude Partner Network, and our AI engineering team builds these jobs for Australian businesses. Our Claude for business page covers how that works, and how to choose who builds Claude into your HubSpot lists what to ask any builder. If you want the HubSpot side first, read how we plugged HubSpot into Claude with MCP.
Not sure whether you need the API or just a plan?
Most teams start with seats and add the API for one repeating job.
Frequently asked questions
How much does the Claude API cost? It is billed per million tokens and depends on the model. Sonnet 5.5 is US$2 in and US$10 out, so summarising a call into HubSpot costs about 1.6 Australian cents.
Is the Claude API different from Claude Pro or Team? Yes. Plans give a person a chat seat at a fixed price, while the API lets your own software call Claude and bills per token.
Can I keep Claude API data in Australia? We could not confirm an Australian processing option on Anthropic’s API. It offers global or US-only inference, and US-only costs 1.1x.
Is there a discount for the Claude API? Yes. The Batch API charges half price for jobs that can wait, and prompt caching cuts the cost of repeated text.
Does the API price include GST? API prices are quoted in US dollars. Our Australian dollar figures are conversions before GST.
Sources
- API pricing, Claude Platform documentation, checked 31 October 2026, for token prices, Batch discount, caching multipliers and the US-only multiplier.
- Data residency, Claude Platform documentation, checked 31 October 2026, for inference and workspace geography options.
- European Central Bank reference rate for 30 October 2026, A$1.4236 per US$1, rounded to A$1.42.
Ready to put Claude to work on a repeating job?
Pick the one task your team repeats most and we will cost it properly.