On 18 June 2026, OpenAI added credit usage analytics and tighter spend controls to ChatGPT Enterprise. The change was reported by Reuters and CIO, and confirmed in OpenAI's own help documentation. If you run an Australian business that has standardised on Claude, the announcement is worth reading for one reason. It confirms that AI spend has moved from an experiment line into a budget that finance now expects to govern. This post looks at what changed, and at how to build cost governance into a Claude rollout before the first surprise invoice lands.
What OpenAI changed on 18 June 2026
The update is aimed squarely at administrators and finance teams. The additions are practical rather than flashy:
Global Admin Console: ChatGPT and Codex credit usage combined in a single view.
Usage by user, product and model: spend can be traced to a team or a workload, not just a total.
Trend tracking: usage over time, including top users and emerging patterns.
A unified Cost API: the same numbers pulled into an organisation's own reporting.
Monthly usage limits: caps set at the workspace, group and individual level.
User credit requests with context: an admin can see why more capacity was asked for.
OpenAI has also said that from 15 July 2026, existing weekly limits will auto-migrate to monthly workspace and group defaults. None of this is unique to OpenAI. It is the direction every serious AI vendor is heading, and the direction your own finance team will push for once Claude usage scales.
Why this matters if you run Claude in Australia
The real signal is not the dashboard. It is that AI cost is now handled like cloud cost: forecast it, attribute it, cap it, and report on it. Australian businesses carry an extra wrinkle, because procurement, data residency and the Privacy Act all shape how you buy and deploy Claude. A Sydney professional-services firm running Claude across 80 staff is making a five-figure monthly commitment, and the CFO will reasonably want to see where it goes.
Put rough numbers on that. At around $90 per seat per month, an 80-person rollout is about $7,200 a month, or close to $86,400 a year, before any usage on the Claude API is added on top. That is a real line item. It deserves the same planning you would give a new piece of cloud infrastructure of the same size.
How Claude cost actually adds up
Claude spend has three moving parts, and most budget surprises come from mixing them up:
Seats: per-user Claude subscriptions across the team. Predictable, and the easiest number to forecast.
API and token usage: anything built on the Claude API, where cost scales with how much text goes in and comes out. This is the part that moves.
Model choice: the Opus, Sonnet and Haiku tiers sit at very different price points. Routing routine work to a smaller model instead of the largest one can cut the bill for that workload by more than half.
Procurement path matters too. You can buy Claude directly from Anthropic, or consume it through AWS Bedrock or Google Vertex AI. It is the same Claude either way, but billing, committed-use discounts and the controls your finance and security teams see will differ. For a business already on AWS with an existing agreement, routing Claude through Bedrock can fold AI spend into a bill the finance team already manages.
The cost-governance layer we build around a Claude rollout
Cost governance should be designed into a Claude deployment in the first week, not added after a billing shock. The practical pieces we put in place look like this:
A usage forecast per team, expressed in AUD per month, so every department starts with a budget rather than an open tab.
Attribution by workload, not just by seat, so a $30,000 quarter can be tied back to the processes it actually served.
Model routing rules, so high-volume, low-stakes tasks run on Haiku or Sonnet and only the hard work reaches Opus.
Alerting thresholds that flag a team at 70 and 90 per cent of its monthly cap, well before month end.
A quarterly review that compares spend against the outcome it was meant to produce, so the number is judged on return, not on size.
Most of this can be assembled from Anthropic's own usage reporting plus a light layer of your own tracking. The point is less about the tooling and more about the habit. Treat Claude spend the way you already treat cloud spend, and the surprises stop.
Claude and ChatGPT on cost controls, fairly compared
OpenAI's June update is a reminder that admin and finance tooling is now part of how these products compete, not just model quality. Anthropic provides usage and cost visibility through the Claude Console and its Team and Enterprise plans, and through the billing views in Bedrock and Vertex for businesses that buy that way. OpenAI's new Global Admin Console and Cost API push hard on the single-pane idea. The honest read in the middle of 2026 is that both vendors are converging on the same set of controls, and the gaps that exist today tend to close within a quarter or two. For an Australian buyer, the deciding factors are usually procurement fit, data handling under the Privacy Act, and which model does your actual work best, rather than one dashboard feature.
Where to start
If Claude is already in your business, the first move is to put a number on it: forecast the next quarter by team, set caps, and decide your model-routing defaults. If you are still planning a rollout, build the budget and the governance into the design now, while it costs nothing to change. We help Australian businesses set up Claude with cost governance in place from day one. If that is on your mind, book a brainstorm with us and we will map it to your teams and your numbers.
