ChatGPT Work pricing explained in one sentence: it is not one plan, it is two products with two separate price tags, and the split is the whole story. If you are trying to work out what it will actually cost your team, you need to know which of the two you are being quoted before you can do the math.
ChatGPT Work pricing explained: two tiers, two bills
OpenAI announced ChatGPT Work on July 9th, 2026, as a workplace product sitting above the consumer plans. For weeks it read like a single offering with a single price. On August 30th, 2026, Simon Willison published a breakdown that laid out what was actually going on: ChatGPT Work is two distinct products bundled under one name, with different pricing and different access.
That distinction matters if you are comparing chatgpt work tiers for a team, because a quote for one tier tells you nothing about the other. Two people at the same company could both say "we pay for ChatGPT Work" and be on completely different plans, with different per-seat costs and different feature access. Get that split wrong in an audit and you size a rollout for the wrong product, or budget for both when only one team actually needs one.
Work Cloud and Work Desktop, what each one actually is
The first product is Work Cloud. It runs through the browser at chatgpt.com or through the mobile apps, the same access pattern as the consumer ChatGPT product, just with workplace-level admin controls and permissions layered on top.
The second is Work Desktop. This is the application that used to ship as Codex, OpenAI's coding-focused agent tool, now folded into the Work branding and sold as a separate desktop product. It is licensed on its own, priced on its own, and a seat on one tier does not include the other.
That is the practical answer to "what does ChatGPT Work cost": it depends which of the two products you are buying, and for how many people. A team that needs Work Cloud for research and writing but not Work Desktop for coding is paying for a different set of seats than a dev team doing the reverse. Understanding chatgpt work tiers means treating these as two line items, not one.
Per-seat pricing means the bill scales with headcount, not with output
Set the exact dollar figures aside for a moment, because the mechanic matters more than the number. Per-seat pricing charges for every person who has a license, regardless of how much that person actually uses it. Someone who opens the tool once a week and someone who runs fifty prompts a day cost the business the same monthly fee.
That is the defining feature of per seat software cost: it is a headcount tax, not a usage fee. Add ten people to a team and the bill goes up by ten seats, whether those ten people generate ten times the output or barely touch the tool. The pricing model has no mechanism for rewarding efficiency. It only tracks who has a login.
What that means once you are past $100k a month
For an ecommerce or service business past $100k a month, ticket volume and workload do not stay flat. Growth means more orders, more support requests, more people needed to keep response times reasonable. Every time you add a support agent or an ops hire to keep up, you also add a seat to whatever per-seat tool that person needs to do their job.
That is the trap. The tool's cost curve is supposed to help you handle growth, but instead it tracks your growth exactly, seat for seat. A support team that grows from four people to nine to handle rising ticket volume does not get a discount for growing, it gets a bigger invoice on the same schedule as its headcount. The per seat software cost model turns operational drag into a recurring, compounding expense instead of removing it.
Rented tool versus owned system, the actual difference
A rented tool is anything billed per seat, per month, for as long as you keep using it. Stop paying and access disappears immediately, along with any configuration, prompts, or workflows you built inside it. You never own the thing, you rent continued access to it.
An owned system is different by design. It gets built once around how your business actually operates, it sits on your own infrastructure, and there is no per-seat rent attached to it. Add ten people to your team and the system does not charge you ten times more, because it was never priced by headcount in the first place. That is the core distinction behind "owned by you": the system keeps working whether your team is 5 people or 50, and you are not paying a subscription to keep the lights on.
Where a tool like ChatGPT Work is the right call
None of this is an argument against per-seat tools categorically. For a small team, or for work that is genuinely light and occasional, a flexible product priced per seat is the right tool. If three people need an assistant for research, drafting, or the odd bit of code, paying per seat costs less than building anything custom and gets them working the same day.
The same is true when the work varies task to task, with no repeatable pattern worth automating. Chatgpt work tiers exist for exactly this kind of general-purpose, unpredictable use. Building a custom system for a job that changes shape every week is a wasted build, because there is no fixed workflow to build around. Rent the tool, use what you need, stop paying if you stop needing it.
Where per-seat pricing breaks down: repetitive ticket volume and handoffs
The picture changes once the work stops being varied and starts being repetitive at volume. A support inbox getting hundreds of near-identical tickets a week, "where's my order," "how do I return this," "my discount code isn't working," is not a per-seat problem, it is a workflow problem. Adding another agent seat to keep response times down just adds another rented license on top of a process that still routes tickets manually between the helpdesk, the order system, and whatever spreadsheet tracks exceptions.
That is where manual work compounds instead of shrinking. Every handoff between tools costs time on top of the seat cost you are already paying, and neither the seat count nor the handoff count goes down as ticket volume rises. This is precisely the gap that ecommerce support automation is built to close: repetitive tickets get handled by a system instead of routed through another paid seat.
How to work out what you are actually paying, seat by seat
You do not need a published statistic to know what a per-seat tool costs you, you need three numbers you already have: price per seat, number of seats, and months you intend to keep paying. Multiply the three and you have the recurring cost of that tool for your team, not the marketing price, the actual one.
Then compare that number to the cost of the manual work the tool is standing in for, meaning the hours your team spends on repetitive tickets and handoffs each month, valued at what those hours actually cost you. If the recurring seat bill plus the manual work still left over costs more than building the workflow once, the math has already answered the question. Rather than guess at a percentage, run your own numbers and work out what the manual work is costing you.
Common questions
Is ChatGPT Work one plan or two?
Two: Work Cloud, accessed through chatgpt.com or the mobile apps, and Work Desktop, the app formerly called Codex, each with separate pricing and access.
Does per-seat pricing get cheaper as a company grows?
No, per-seat pricing multiplies with headcount by design, so more seats always means more rent, regardless of tier.
When does an owned system make more sense than a per-seat tool?
Once ticket volume or team size is high enough that seat costs and manual handoffs cost more than building the workflow once.
Can ChatGPT Work replace ecommerce support automation?
It can help a person do the work faster, but it does not remove the per-seat cost or the manual handoffs between your helpdesk and your other systems.
If your support team is growing seats faster than it's closing tickets, look at the handoffs before you look at the subscription. We map the current workflow and build what replaces it, including for teams weighing up a cheaper helpdesk, which still doesn't touch the per-seat problem. The mechanics are on our ecommerce support automation page.
