Pricing figures updated 6 September 2026: USD prices are converted at the current rate (1 USD ≈ 1.40 AUD, verified 5 September 2026) and shown ex-GST, replacing the retired 1.55 + GST basis. A GST-registered Australian business that supplies its ABN is not charged GST on an imported digital service; other buyers should allow ~10% on top (Methodology).
Archived edition: our verdict as at 23 Jun 2026. Kept unchanged as the record of what we said then. Read the current edition: The End of the $20-a-Month Era: AI Pricing Is Splitting in Two for Australians · Why we keep every edition
By Adam, Editor
For three years, the price of serious AI was twenty dollars a month. ChatGPT Plus, Claude Pro, Gemini Advanced, Cursor, the lot — all landed within a dollar or two of US$20. You paid your seat, you got the best model the company sold, and you used it as much as you liked. Simple, flat, unlimited enough.
That deal is over. It did not end with an announcement. It ended the way these things usually do — quietly, in the billing pages, over a few weeks in June.
And it did not end in one direction. The twenty-dollar seat is being pulled apart from both ends at once. At the agent end, the flat fee is becoming a fee plus a meter. At the assistant end, the flat fee is becoming too expensive — the floor is falling toward eight dollars. Either way, the mental model most of us still carry — one price, all you can eat — is now wrong. For an Australian paying in a weaker currency, getting that model right is worth real money.
What changed at the top end: the meter arrived
Start with the coding and agent tools, because that is where the shift bites hardest.
On 1 June 2026, GitHub Copilot moved every plan onto usage-based GitHub AI Credits. The headline seat prices did not move — Pro is still US$10 a month — and ordinary code completions stay free and unmetered. But the old safety net, where the tool quietly dropped to a cheaper model and kept working once you hit a limit, is gone. Run long, autonomous agent sessions and you now draw down a credit balance; when it's empty, you pay on demand or you stop (our breakdown).
A day later, on 2 June, Windsurf relaunched as Devin Desktop under Cognition — and the product Australians now download bills the same way: a usage quota on the paid plans, with overage charged at API rates once you exhaust it (review). Its sibling, the autonomous Devin cloud agent, is metered to the core (review). Cursor got there earlier still: since mid-2025 each paid plan has carried a pool of model usage roughly equal to the plan price, and picking a premium frontier model on every request drains that pool fast, after which the on-demand charges start (review).
None of these is a price rise, exactly. The seat still costs what it cost. But the seat no longer buys the work — it buys an allowance, and the work is metered underneath it.
There's a real engineering reason for this, and it's worth understanding because it isn't going to reverse. A chatbot answering a question is cheap. An agent — one that reads your whole codebase, plans a change across a dozen files, runs the tests, reads the errors and tries again — can burn through hundreds of times more compute in a single task. A flat US$20 covers the first. It cannot cover the second without either a meter or a queue. The industry chose the meter. As AI tools shift from answering to doing, expect metering to follow them out of the coding category and into everything else.
What changed at the bottom end: the price war
Now look the other way, at the everyday assistant — the tool a Melbourne accountant or a Brisbane tradie actually buys to help with email, documents and research.
Here the pressure runs the opposite way: down. On 11 June, Google cut Gemini AI Plus in Australia from A$12.49 to A$7.99 a month — a 36% drop, as a standard price, not a teaser (the cut). At eight dollars, billed natively in Australian dollars and bundled into Gmail, Docs and Sheets, the cheapest serious paid AI assistant just got cheaper than a weekly coffee.
That is not a coincidence sitting next to the metering story. It's the same story from the other side. When the genuinely expensive work moves to a meter, the commodity work — drafting, summarising, answering — gets cheap enough to give away as a loss-leader. Google can afford A$7.99 because the assistant is the on-ramp, not the profit centre. The twenty-dollar flat tier is being squeezed from below by tools that do the everyday 80% for less than half the price.
Why this lands harder in Australia
Two reasons an Australian buyer should care more than an American one.
First, most of the metered tools bill in US dollars, with GST added at checkout. A US$20 seat is already closer to A$28 ex-GST once you account for the exchange rate and tax. Add a meter on top and your "A$28 a month" becomes "A$28 a month plus whatever the agent spent" — and that second number is the one you can't predict in advance. Variable USD billing is the worst kind of cost for a small business trying to forecast: foreign, taxed, and uncapped.
Second, the cheapest, most predictable option is now the local one. Google's A$7.99 is set in Australian dollars by Google — no FX, no guesswork, GST handled. For once, the lowest-friction choice and the lowest-cost choice are the same choice, and it's the one priced for a local card. That hasn't been true in this market before.
How to pay for AI now
The flat-seat era had one virtue: you didn't have to think. That's gone, so here is the thinking it's replaced with. Split your AI spend into two buckets, because they now price on opposite logic.
The assistant bucket — email, documents, research, the daily help. This is getting cheaper and the differences between tools are narrowing. Buy on price, currency and integration, not on raw capability. For most Australians that points at Gemini at A$7.99 inside Workspace, or whichever assistant already lives in the tools you use. Don't overpay for an assistant; that's the part the price war is handing you for free.
The agent bucket — coding, multi-step automation, anything that acts rather than answers. Here the seat price is the entry fee, not the bill. Before you commit, model your likely usage, turn on every budget cap and spend alert the tool offers, and read the meter for the first month before you trust the headline number. The right question is no longer "what's the monthly price" but "what does a heavy week actually cost, and can I cap it." Treat the metered tools the way you'd treat a work phone on global roaming: useful, worth it, and capable of a nasty bill if you don't set limits.
The twenty-dollar seat was a simplification we all got comfortable with. Its end is genuinely a loss — predictability is worth something. But it's also a clarifying moment. The cheap, commodity AI is getting cheaper, and the genuinely powerful, compute-hungry AI is finally being priced like the resource it is. Pay for each on its own terms, and you'll spend less and forecast better than the buyer still reaching for a single flat number that no longer exists.
We score every tool against the same five criteria, including Australian Fit and how it bills — see our Methodology, the best AI coding tools roundup for the metered end, and the best AI assistants roundup for the bargain end.
Sources
- GitHub blog — Copilot moving to usage-based billing (primary): https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/ (effective 1 June 2026)
- Devin / Cognition — "Windsurf is now Devin Desktop" (primary): https://devin.ai/blog/windsurf-is-now-devin-desktop/ (2 June 2026)
- Cursor pricing (primary): https://cursor.com/pricing
- Google Gemini AU subscriptions (primary): https://gemini.google/au/subscriptions/?hl=en-AU (AI Plus A$7.99, effective 11 June 2026)
- Cross-references: AI Directory reviews of GitHub Copilot, Cursor, Devin Desktop and Devin; news items on the Copilot billing change and the Gemini price cut. Figures anchored to the dossiers and reviews behind each.