The Month in Australian AI: April 2026

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).

April was the month the price tags moved. Two things happened at once, and they pulled in opposite directions for Australian buyers.

First, headline seat prices fell. Cheaper plans, cheaper agent runs, more in the free tier. On paper, a good month to be paying in Australian dollars.

Second, "agentic" stopped being a premium add-on and became the default. The tools you already pay for now ship agents that act on your behalf — write code, run tasks, fill in slides. That is powerful. It also quietly changes what you are billed for.

The net effect for an Australian small business: lower entry prices, but a real risk that usage-based costs creep up once the agents are doing the work. Here is how the month played out.

ChatGPT dropped the Business seat to US$20 — annual

On 2 April, OpenAI cut the ChatGPT Business seat to US$20 (~A$28 ex-GST) per user — but on annual billing. It also introduced Codex-only seats for teams that just want the coding agent.

For Australians, the annual catch matters. US$20 a seat is genuinely competitive. But it is a yearly commitment, not the month-to-month flexibility a lot of small operators prefer. Run the maths before you lock in eleven seats you might not use in October.

Notion, Cursor and the agent price war

The same day, Cursor shipped Cursor 3 with parallel agents — multiple coding agents working at once. On 14 April, Notion 3.4 Part 2 landed Custom Agents and made agent runs 35–50% cheaper.

This is the April story in miniature. Vendors are racing to make agents cheaper per run so you use more of them. Cheaper per unit, more units. For a Melbourne marketing team, that can be a bargain or a bill shock, depending on whether anyone is watching the meter.

Open weights kept coming: Gemma 4

On 2 April, Google released Gemma 4 with open weights — a model you can download and run yourself.

For most Australian businesses this is not a "switch today" moment. But it matters for one reason: data residency. If you can run a capable model on your own infrastructure, the question of where your data sits stops being someone else's policy. For finance, legal and health operators, that is worth tracking.

Governance got open-sourced

Also on 2 April, Microsoft open-sourced its Agent Governance Toolkit — tooling for controlling what agents are allowed to do.

The timing is not an accident. The moment agents act on your behalf, you need rules about what they can touch. For an Australian business handling customer data under the Privacy Act, "what is this agent permitted to do" is a question you want answered before, not after.

The reminder that agents can go wrong

April also delivered two cautionary tales. On 6 April, Linear published a post-mortem on a 24 March permission-filter incident. On 22 April, Lovable responded to its own April incident.

The lesson is plain. As agents do more, the blast radius of a bug gets bigger. Read the post-mortems. A vendor that publishes one honestly is doing better than one that stays quiet.

More in the free tier

April was generous at the bottom of the market. Google moved Vids to a free tier. OpenAI shipped ChatGPT Images 2.0 on 21 April. Anthropic added Claude Code Routines and a desktop redesign on 14 April, and Claude Opus 4.7 arrived around 17 April. Perplexity launched Personal Computer on Mac, also around 17 April. Late in the month, Lovable shipped a mobile app.

For an everyday Australian wanting to try AI without a credit card, there has never been more in the free tier. Google Vids in particular gives small businesses a no-cost way to make explainer and training videos.

The one to read carefully: GitHub Copilot

If you use GitHub Copilot, April had two items you cannot skim.

On 24 April, GitHub confirmed Copilot data is used to train its models unless you opt out — across Free, Pro and Pro+. Then on 27 April, GitHub announced usage-based billing, effective 1 June.

Both have an Australian edge. The training default means your code goes into the machine unless you actively turn it off — check your settings. The billing change means a predictable monthly cost becomes a variable one from June. For a solo developer or small studio budgeting in AUD, that is the difference between a known line item and a surprise.

What it meant for Australians

April handed Australians two things and asked them to hold both at once.

The good: cheaper seats, cheaper agent runs, and more capability in the free tier than ever. If you have been waiting for AI to get affordable, this was your month.

The catch: the pricing is shifting from flat seats toward usage. Agents are now the default, and agents cost money every time they run. The headline price fell; the metered price is the one to watch.

Three actions before May. Check whether Copilot is training on your code, and opt out if you want to. Pencil in 1 June for Copilot's billing change. And if you are turning on agents, turn on a budget alert at the same time. The cheap month only stays cheap if someone is watching the meter.

Sources

ChatGPT Business pricing — OpenAI (Primary), checked 30 April 2026: https://openai.com/chatgpt/business/ — Wayback snapshot queued (SPN pending)
Notion 3.4 — Notion (Primary), checked 30 April 2026: https://www.notion.so/releases — Wayback snapshot queued (SPN pending)
GitHub Copilot data and training settings — GitHub (Primary), checked 30 April 2026: https://docs.github.com/en/copilot — Wayback snapshot queued (SPN pending)
GitHub Copilot usage-based billing — GitHub (Primary), checked 30 April 2026: https://github.blog/ — Wayback snapshot queued (SPN pending)