Chatgpt API Pricing in Australia (2026)

For a small service business, the OpenAI token bill can be only a few Australian dollars a month. People search for the “ChatGPT API”, but the product is the OpenAI API, billed separately from ChatGPT subscriptions. The bigger budgeting question is what the connected system costs to build and keep running. My guide to AI for small business puts that decision in context.

What Does the Openai API Cost per Million Tokens In 2026?

The table below uses standard rates checked on 21 October 2026. OpenAI shows prices with a $ sign; I have treated those as US dollars as published. The Australian figures are my approximate conversions, not official AUD prices.

The RBA recorded US$0.6984 per A$1 on 20 October 2026, making US$1 approximately A$1.4318. Your card or bank rate will differ, and bank fees can change the final amount.

Input is the material sent to the model; output is the material generated. Budget for both. These are OpenAI's standard rates; long prompts can cost more. Reasoning tokens can also be billed as output under OpenAI's pricing rules.

Which Openai Model Should a Small Service Business Use?

As at 21 October 2026, OpenAI describes GPT-6 Astra as its most capable model for demanding work, GPT-6.1 Sol as offering near-Astra performance at a lower cost, and GPT-6 Luna as its most efficient model for focused, high-volume tasks.

OpenAI's default starting suggestion is Astra. I would test Luna and Sol against actual enquiries before paying Astra rates for routine admin.

• Try Luna for sorting enquiries and drafting tightly specified standard replies.
• Test Sol when replies need more context and judgement.
• Compare Astra on difficult examples where cheaper models miss something material.

Judge the drafts against your business rules, including when the system should stop and ask a person. My business automation model comparison covers that choice more broadly.

What Would 500 Enquiry Reply Drafts Cost Each Month?

Here is an illustrative example: 500 drafts a month, each using 3,000 input tokens for the enquiry and business context, plus 500 output tokens. That totals 1.5 million input and 250,000 output tokens.

The example assumes one generation per draft, no cached-input discount and no additional reasoning tokens or tool calls. Retries, longer conversations and extra checking calls change the total.

So I would spend more attention on whether the draft uses the right information, reaches the right inbox and waits for approval. Saving a few dollars on tokens does little if the workflow creates more checking work.

How Do Batch, Flex, Fast Mode and Caching Change the Bill?

These options change the rates or the amount of input charged at the full rate. Terms below are current as at 21 October 2026.

• Batch: 50% cheaper than synchronous APIs, with batches completing within 24 hours. I would consider it for overnight CRM tidying or scheduled summaries.
• Flex: Batch token rates, with possible caching discounts. Capacity is not guaranteed; insufficient resources return an uncharged 429 error. I would reserve Flex for work that can wait.
• Fast mode: the three flagship rows cost twice their standard rates in the pricing table. Priority processing was renamed Fast mode on 30 July 2026.
• Prompt caching: enabled by default on supported models. For GPT-5.6 and later, the minimum cacheable prompt is 1,024 tokens. OpenAI describes cached-input discounts of up to 95%; actual savings depend on repeated content.

A stable block of business instructions at the start of each request is the kind of repeated prefix caching rewards. I would keep live customer replies on standard processing and test discounts on less urgent work first.

How Is the API Billed, and Do Chatgpt Subscriptions Cover Usage?

ChatGPT subscriptions and API usage are separate bills. API usage is pay as you go against credits bought in advance. Check the billing dashboard and set a spending limit before connecting a live workflow.

As at 21 October 2026, OpenAI's usage tiers upgrade automatically as total credit purchases reach thresholds; higher tiers generally provide higher rate limits. A usage allowance is a ceiling, not a monthly fee or a suggested budget.

The published pricing page has no tax note. Check your invoice and ask your accountant how GST applies to an overseas digital service. The AUD amount moves with the exchange rate and any bank fees.

What Extra Costs Sit on Top of Tokens?

Australian data residency needs careful reading. As at 21 October 2026, OpenAI lists Australian regional storage without Australian regional processing. Access requires Modified Abuse Monitoring or Zero Data Retention approval and additional requirements.

By default, API data is not used to train OpenAI models unless the customer opts in. Default abuse monitoring logs can still be retained for up to 30 days, according to the same guide. Ask what is stored and processed where before choosing a configuration.

Your budget also needs the CRM, inbox and automation platform subscriptions being connected. See the Australian AI automation cost guide for the wider budget, or Claude pricing in Australia when comparing model options.

What Should a Build and Ongoing Running Quote Cover?

I would ask for three separate costs: the initial build, ongoing software and API usage, and ongoing running. A cheap model does not remove the work of connecting systems, testing drafts or handling failures.

1. Define the workflow. Name the trigger, source information, proposed action and approval point.
2. Show the usage assumptions. Request volumes, input and output lengths, model choices, retries and tool calls.
3. Specify the build. Include connections, prompt testing against real examples and the human approval step.
4. Define ongoing responsibility. State who monitors failures, reviews exceptions and updates the system when models or prices change.
5. Show the bills. Identify account ownership, software subscriptions, spending limits and any usage markup.

For an Australian service business that wants automation designed, built and then run, with a person approving anything sensitive, Truespeak is built for exactly that.

At Nikon Australia I built an AI chatbot that deflected more than 140 customer service calls a month.

I work directly with clients and match the model to the job. Truespeak works with existing tools; where a tool has no usable connection, that step stays manual. Truespeak's managed AI operations cover monitoring, fixing failures, reviewing exceptions and improving rules after launch.

Truespeak quotes after a discovery call. The quote covers design, build and ongoing running.