Estimated reading time: 8 minutes · Prices last checked: 10 October 2026
AI token pricing works differently from a subscription. AI APIs don’t charge a monthly fee; they charge for every token you send and every token you get back, in US dollars. This guide shows what the main models from OpenAI, Anthropic (Claude) and Google (Gemini) cost per million tokens in October 2026, converted to Australian dollars. It also works through what a real job costs and the quickest ways to cut the bill.
Short answer: the cheapest current models, GPT-6 Luna and Claude Haiku 5.5, cost about A$0.14 per million input tokens and A$0.72 per million output tokens. The top models, GPT-6 Astra and Claude Fable 5.1, cost about A$14.33 in and A$71.64 out. A million tokens is roughly 750,000 words. Output always costs more than input: 5 to 8 times as much at today’s list prices.
Jump to: What a token is · How pricing works · Prices in AUD · A real job · How to pay less · Aussie gotchas · FAQ
What is a token?
A token is a chunk of text: a short word, part of a longer word, or a punctuation mark. AI models read and write in tokens, and tokens are what you’re billed for. OpenAI’s rule of thumb for English is that one token is about four characters, or three-quarters of a word, so 100 tokens is about 75 words (OpenAI). Google gives a similar figure for Gemini: about four characters per token, or 60 to 80 English words per 100 tokens (Google).
You pay for two kinds of token:
- Input tokens are everything you send: your question, your instructions, any document you paste in and, in a chat, the whole conversation so far.
- Output tokens are everything the model writes back. Google bills a model’s hidden thinking as output too (Google).
Token counts aren’t the same everywhere. Anthropic says the tokenizer in Claude 4.7 and later models produces about 30% more tokens for the same text than its older one (Anthropic). So two models with the same price per token can cost different amounts for the same job. Test with your own text before you compare.
How token pricing works
Cost = (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price)
Prices are quoted per million tokens. Three things change what you actually pay:
- Cached input is much cheaper. If you send the same text again, such as a long set of instructions, the repeated part is billed at a fraction of the normal rate. On Claude Sonnet 5.5 and GPT-6.1 Sol, cached input costs US$0.10 per million tokens instead of US$2.00. Anthropic charges 1.25 times the normal input price to write text into a 5-minute cache.
- Batch jobs are half price. OpenAI, Anthropic and Google all take 50% off input and output if you send work as a batch and wait for the results. OpenAI says batch jobs run over 24 hours (OpenAI).
- Very long prompts can cost more per token. Gemini 3.1 Pro doubles its input price above 200,000 tokens, and Claude Haiku 5.5 charges five times more above 100,000 tokens. OpenAI lists a separate, higher long-context rate. Anthropic’s other current models charge the same rate across their full 1-million-token context.
AI API prices in USD and AUD (October 2026)
Standard prices per million tokens, from each provider’s official pricing page. AUD is converted at the Reserve Bank rate for 9 October 2026 (US$1 = A$1.4329).
| Model | Provider | Input per million | Output per million | Notes |
|---|---|---|---|---|
| GPT-6 Astra | OpenAI | US$10.00 (A$14.33) | US$50.00 (A$71.64) | Top tier |
| Claude Fable 5.1 | Anthropic | US$10.00 (A$14.33) | US$50.00 (A$71.64) | Top tier |
| Claude Opus 5.5 | Anthropic | US$4.00 (A$5.73) | US$20.00 (A$28.66) | Upper mid tier |
| Gemini 3.1 Pro | US$2.00 (A$2.87) | US$12.00 (A$17.19) | Still labelled Preview; prompts up to 200,000 tokens | |
| GPT-6.1 Sol | OpenAI | US$2.00 (A$2.87) | US$10.00 (A$14.33) | Mid tier |
| Claude Sonnet 5.5 | Anthropic | US$2.00 (A$2.87) | US$10.00 (A$14.33) | Mid tier |
| Gemini 3.8 Flash | US$0.75 (A$1.07) | US$3.75 (A$5.37) | Price doubles on 1 January 2027 | |
| Gemini 3.5 Flash-Lite | US$0.30 (A$0.43) | US$2.50 (A$3.58) | Budget | |
| GPT-6 Luna | OpenAI | US$0.10 (A$0.14) | US$0.50 (A$0.72) | Budget |
| Claude Haiku 5.5 | Anthropic | US$0.10 (A$0.14) | US$0.50 (A$0.72) | Budget; prompts up to 100,000 tokens |
Heads-up: Google’s price for Gemini 3.8 Flash rises on 1 January 2027, from US$0.75 to US$1.50 per million input tokens and from US$3.75 to US$7.50 per million output tokens (Google). If you’ve built something on it, budget for double.
Prices are from OpenAI (standard, short-context rates), Anthropic and Google (paid tier). You’re billed in US dollars, so what lands on your card will differ from the AUD column (see Aussie gotchas).
What a real job costs
Say you want to summarise 1,000 articles of about 3,000 words each into 300-word summaries. Using the rule of thumb above, each article is about 4,200 input tokens (the article plus your instructions) and 400 output tokens. Across 1,000 articles that’s 4.2 million input tokens and 0.4 million output tokens.
| Model | Cost for 1,000 summaries | As a batch (50% off) |
|---|---|---|
| GPT-6 Astra or Claude Fable 5.1 | A$88.84 | A$44.42 |
| Claude Opus 5.5 | A$35.54 | A$17.77 |
| Gemini 3.1 Pro | A$18.91 | A$9.46 |
| GPT-6.1 Sol or Claude Sonnet 5.5 | A$17.77 | A$8.88 |
| Gemini 3.8 Flash | A$6.66 | A$3.33 |
| Gemini 3.5 Flash-Lite | A$3.24 | A$1.62 |
| GPT-6 Luna or Claude Haiku 5.5 | A$0.89 | A$0.44 |
The same job costs under A$1 on a budget model and almost A$89 on a top model. For simple work like summarising, start with a budget model and only move up if the results aren’t good enough. These figures use the rule-of-thumb token counts, so treat them as estimates. Real counts vary by model.
Seven ways to pay less
- Start with the cheapest model that does the job. At list prices, budget models cost 1% of what top models cost per token.
- Batch anything that can wait. Overnight reports, bulk tagging and document processing all qualify for the 50% batch discount.
- Put repeated text first, so it can be cached. Caching works on the repeated beginning of a prompt, so keep your long, unchanging instructions at the start and the changing part at the end. On Claude Sonnet 5.5, cached input is 95% cheaper.
- Cap the output. Output costs 5 to 8 times as much as input. Ask for short answers and set a maximum output length in your app or tool.
- Keep conversations short. In a chat, every new message resends the whole conversation as input. Start fresh, or summarise the history, once a thread gets long.
- Stay under the long-prompt limits. Splitting a document can keep you under Gemini 3.1 Pro’s 200,000-token limit or Claude Haiku 5.5’s 100,000-token limit, where the price per token jumps.
- Test on a free tier, but not with private data. Google offers a free tier for its Flash and Flash-Lite models, but says free-tier data is used to improve its products. Paid-tier data isn’t (Google).
Gotchas for Australians
- You’re billed in US dollars. The AUD figures above move with the exchange rate. Your bank may also add an international transaction fee of 3% to 3.5%; How to pay for ChatGPT and Claude without bank fees explains how to avoid it.
- GST may be added. Overseas companies have to charge GST to Australian consumers (ATO), but business accounts with an ABN can be invoiced differently (Sandlabs). Check your invoice, and see Does ChatGPT charge GST in Australia? for the details.
- Your ChatGPT subscription doesn’t cover the API. OpenAI says its APIs are billed separately from ChatGPT Plus, Business, Enterprise and Edu (OpenAI).
- Data residency costs extra. OpenAI charges 10% more for its regional processing (data residency) endpoints on models released from 5 March 2026 (OpenAI).
- Prices change often. Every model in the original version of this post has since been replaced in this table, and Google has already announced a price rise for January 2027. I re-check these prices monthly.
Subscription or API?
If you’re one person chatting with an AI, a subscription is usually simpler: a flat A$30 to A$34 a month for ChatGPT Plus, Claude Pro or Google AI Pro, compared in Claude vs ChatGPT vs Gemini in AUD. The API makes sense when you’re building something, such as an app, an automation or a pipeline that processes lots of documents, and you want to pay only for what you use. For every AI subscription priced in Australian dollars, see The Real Cost of AI in Australia.
FAQ
What is an AI token?
A chunk of text the model reads or writes. In English, one token is about four characters or three-quarters of a word, so 1,000 tokens is roughly 750 words.
How does token pricing work?
You pay a price per million input tokens (what you send) plus a price per million output tokens (what the model writes). Output costs 5 to 8 times as much as input on today’s main models. Cached input and batch jobs cost less.
Which AI API is cheapest in 2026?
At list price, GPT-6 Luna and Claude Haiku 5.5 tie at US$0.10 per million input tokens and US$0.50 per million output tokens (about A$0.14 and A$0.72). Google’s Flash and Flash-Lite models have a free tier for testing.
Does ChatGPT Plus include API access?
No. OpenAI bills API usage separately from ChatGPT Plus, Business, Enterprise and Edu.
Do I pay GST on AI API usage in Australia?
Usually, if you’re a consumer: overseas suppliers have to charge GST to Australian consumers. Business accounts with an ABN can be invoiced differently, so check your invoice or ask your accountant.
How can I reduce token usage?
Use the smallest model that works, batch work that can wait, keep repeated instructions at the start of the prompt so they’re cached, cap output length and keep conversations short. See Seven ways to pay less.
Related guides on FNG
Written by Scott (Freaky Newguy), a Sydney-based IT worker. How I check prices.
Sources
- API prices: OpenAI API pricing, Anthropic Claude API pricing, Google Gemini API pricing (all checked 10 October 2026)
- Batch timing and ChatGPT vs API billing: OpenAI pricing FAQ
- Tokens per word: OpenAI, What are tokens; Google, Understand and count tokens
- Exchange rate: Reserve Bank of Australia exchange rates
- GST on imported digital services: Australian Taxation Office; Anthropic and ABNs: Sandlabs