Most owners I speak to have been handed the same AI plan by their IT provider: Copilot, because it is Microsoft. Very few have seen it next to the alternatives. I have used all three for a year on real work, and in August the first Cowork bills landed, my own seat over $400 for the month. So this is both halves on one page: what comes back, and what it costs per person, in Australian dollars.
What one person costs a month, by what they do
Published prices, AU$, ex GST, annual billing. Bold is the cheaper buy at that level.
| AU$ a person a month | Copilot | ChatGPT | Claude |
|---|---|---|---|
| Light · Chat“Draft the reply. Summarise this.”0.3% of a $110k salary | $27Promo, $31 after | $28Business seat | $28Team seat |
| Moderate · Cowork“Reconcile it, build the deck.”1.5% of a $110k salary | $193–442Seat + credits | $139Flat, capped | $139Flat, capped |
| Heavy · Code“Build us an app that reads Xero.”3–4.5% of a $110k salary | $305–580Copilot Studio credits | $277–416Codex + US$100–200 | $277–416Claude Code + US$100–200 |
Swipe sideways to see the whole table.
Published prices as at 3 September 2026, US dollars at 0.72. Moderate on Copilot is the seat plus credits at one US cent each, 200 to 500 a task. Heavy on Copilot is Copilot Studio: the seat plus one or two prepaid packs of 25,000 credits at US$200. Your existing Microsoft 365 licence is excluded from all three.
What to budget, by role, as a share of salary
Pale is the seat licence alone; blue is the seat plus heavy usage.
| Role | AU$ a month | Share of salary |
|---|---|---|
| Admin assistantChat | $28On $70k | |
| AccountantChat, then Cowork | $28–139On $100k | |
| Graphic designerChat + image credits | $60–120On $85k | |
| Researcher, analystCowork | $139–280On $120k | |
| Software developerCode | $350–500On $150k |
Swipe sideways to see the whole table.
Salaries are mid-market Australian figures for a firm of this size. Developer usage is the US$150–250 a month Anthropic publishes for Claude Code, on top of the seat. Designer adds an image or video plan to a chat seat. The Ramp AI Index puts the median US firm at about US$11 an employee a month and the top 10% at US$650, so 1–4% of salary sits inside the range of serious adopters.
1It answers a question. The others do the job.
Every comparison you will be shown is a features list, and on features they look alike. The difference is what comes back. Hand Claude a whole problem, “work out why margin dropped in Q3 and draft the note to the board”, and it splits the job, runs the parts at once and returns something you can send. It will open your browser and read the supplier portal you are logged into. Copilot answers the question you typed, inside the app you are in. While your team asks it to summarise an email, the firm down the road has handed an agent the whole board pack and gone to lunch. That gap compounds every month and never appears on an invoice.
2It is slower, and Claude in Copilot isn’t the same.
Ask all three the same thing and Copilot is the one you wait for. Not long enough to complain, long enough to put the kettle on, and by Friday that is a lot of cups of tea across a 25-person firm. You will also see Opus and GPT in Copilot’s model dropdown and assume you have Claude and ChatGPT. You do not. You have the engine, minus everything the other two built around it: the agents, the browser, the long documents. Claude holds roughly a million words in view at once, about 2,000 pages, and ChatGPT a few hundred pages. Copilot inside Word or Outlook works from a fraction of that, which is why it reads the start of a contract and skims the end. It is a Ferrari badge on the family Camry.
3Past chat, Copilot is the dear one.
At the chat level the three cost the same, and Copilot is arguably the best value because it sits inside the tools your people already have open. The moment someone hands work to an agent, the pricing splits. Claude and ChatGPT sell a premium seat at about $139 a month, flat and capped. Copilot keeps the $27 seat and meters every agent run in credits, so the same person costs $190 to $440 and you learn which in arrears. For a builder it is Copilot Studio, $305 to $580 with the seat against $277 to $416 elsewhere, and for the higher price you get low-code agents inside the Microsoft stack, not the standalone app that reads Xero and emails the partners on Friday. Flat and capped versus cheap and metered is the real choice, and the crossover is around one agent task a day.
4$139 is the floor, not the ceiling.
A premium seat is not unlimited. It is an allowance: a rolling five-hour window plus a weekly cap, shared across chat, Cowork and Code, and the vendors adjust it. Anthropic’s Claude Code documentation puts a developer’s usage at US$150 to US$250 a month, more than the seat itself, and Microsoft reportedly pulled its engineers off Claude Code in June after individuals hit US$500 to US$2,000 a month. A researcher who hands Cowork a folder three times a day will feel the weekly cap by Thursday. Budget the seat as the floor and expect the doers to need top-ups.
5Budget by person, not by product.
Against a $110,000 salary the three seats cost 0.3%, about 1.5%, and 3 to 4.5% respectively. Most firms of ten are seven chatters, two doers and one builder, which comes to roughly $900 a month, or about 1% of payroll. That is my rule of thumb: 1% of payroll, split across three kinds of seat, reviewed every six months as people move up the ladder. The mistake is buying ten identical seats and then being surprised that the builder is throttled and the seven chatters never log in.
6Your IT provider probably likes Copilot.
Since July, Microsoft pays its resellers more when a client takes Copilot, and the guidance to partners this year is to “lead with Copilot bundles” for small business. Your IT provider runs your tenant, so the tool inside it is less work for them, and if it disappoints, it was Microsoft’s product. My advice: keep Copilot if you already have it, for meeting notes and finding that email. Put the real work, the documents, the models, the decks, on Claude or ChatGPT, starting with the five people whose output matters most. Then ask your IT provider what they would recommend if Microsoft paid them nothing, and watch the pause.
Sources A year of side-by-side use on Scenario Three work · Gartner Copilot customer surveys and NPS reporting, 2025–26 · Microsoft partner incentive announcements, July 2026 · Microsoft Copilot, Anthropic and OpenAI published documentation on document limits, context windows, pricing and credit billing · Anthropic Claude Code cost documentation · Windows Central reporting on Microsoft engineer usage, June 2026 · Ramp AI Index, July 2026 · My own Cowork and Claude Code bills, August 2026
General information only, based on published list prices and my own use of these tools. Prices, plan names and exchange rates move often; check the vendor’s current pricing and your own Microsoft agreement before acting. Product names are the property of their owners.