The short version: OpenAI added a second seat tier to ChatGPT Business pricing on August 10. Premium seats run $125 per user per month, or $100 per user per month billed annually, against $25 and $20 for the Standard seats you already have. Premium buys five times the usage and lifts the five-hour cap. Most small teams do not need it across the board, and you can mix the two tiers in one workspace, so the useful question is not whether to upgrade but which specific person to upgrade.
The interesting part is not the number. It is the sequence. OpenAI lifted the five-hour usage limit on Business seats on July 12, restored it on July 31, and on August 10 turned removing it into a $125 product, as CIO documented in its timeline of the change. A ceiling that moves three times in a month is not a hardware constraint. It is a pricing boundary, and it has now been given a price.
What actually changed in ChatGPT Business pricing?
One tier became two. Standard seats stay where they were, at $25 per user per month or $20 billed annually. Premium seats sit on top at $125 per user per month or $100 billed annually, and OpenAI describes them as giving your most active people five times the usage of a Standard seat without the five-hour limit. Usage resets weekly on both tiers.
Two details matter more than the headline price. First, administrators can mix seat types inside a single workspace and reassign them as workloads shift, so this is not an all-or-nothing upgrade. Second, OpenAI is running an early-access promotion of $100 in workspace credits, listed as 2,500 credits, for each Premium seat added, up to five seats.
Treat that promotion carefully. A credit offer with a short window is designed to pull the decision forward, and a deadline is a poor reason to change a recurring bill. The seats will still be there in September.
Why would OpenAI charge five times as much for the same chatbot?
Because it is no longer the same chatbot. A person typing questions into ChatGPT consumes a short burst of computing and then stops. A person handing an agent a multi-step job walks away while the agent keeps working, reading, retrying, and checking itself for as long as the task takes. The Decoder frames the split around exactly that shift, noting that agent-driven work consumes far more capacity than conversation does.
An analyst quoted by CIO put the structural point better than the press release did: this is not vendors admitting that flat seats failed, it is vendors admitting that one seat no longer describes one economic profile. That sentence is worth sitting with, because it applies to every AI tool you buy, not just this one.
For most of the last fifteen years, per-seat pricing worked because people used software at roughly comparable rates. Your bookkeeper and your estimator both opened the same accounting package a similar number of times. Agentic AI breaks that assumption. One person running long automated jobs can consume more capacity in a week than five colleagues do in a month, and the flat seat quietly subsidised them until somebody noticed.
Who on your team actually needs a $125 seat?
Whoever is hitting the cap. That sounds obvious, and it is the whole method, because the cap is now the measurement.
Here is the part that is genuinely new and easy to miss. Until this month, you had no per-person read on how hard anyone in your business was leaning on AI. Everyone cost the same, so the invoice told you nothing. Now the people who need Premium identify themselves by running out of room, and that is management information you did not previously have. It tells you where the heavy, repetitive, high-volume work in your business has quietly collected, and it usually is not where the org chart suggests.
So before upgrading anybody, spend two weeks watching for one signal: who reports getting cut off mid-task. If nobody does, your team is a Standard-seat team and this announcement costs you nothing. If one person does, upgrade that one person and leave everyone else alone. If half your team does, you have learned something more valuable than a seat decision, which is that a large amount of your operational work is now being absorbed by AI under human review, and that is worth planning around deliberately.
Worth being clear about what the Premium seat is for. It is capacity for a person who is already doing the work, so a longer job finishes in one sitting instead of stopping halfway. It buys the person fewer interruptions, not fewer people.
What does this do to the rest of your AI budget?
It pushes in the opposite direction from the trend we covered a few weeks ago. Raw model prices have been falling, and we wrote about that in the AI model price war and what it changes for small business software. Seat prices are now rising at the top end. Both things are true at once, and they are not in conflict. The intelligence is getting cheaper per unit while the people using the most of it are being asked to pay separately.
Set against the alternatives, a fixed $125 ceiling is not the worst structure on offer. Pure metered billing is, and small businesses have already been burned by it: we covered how agentic workflows pushed one GitHub Copilot bill from $39 to $847 in the Copilot metered billing shock. A predictable seat you can budget beats a meter you cannot. Microsoft, meanwhile, went the third way and folded Copilot into the base plans, which we broke down in the Microsoft 365 Copilot break-even math.
Three vendors, three answers to the same problem, all within about a month. Tiered seats, metered consumption, and bundling. That is what an unsettled market looks like, and it is the practical reason to stay on monthly billing a little longer than feels comfortable. The annual discount on a Premium seat saves you $300 a year per person and locks you into a pricing model that is visibly still being figured out.
If none of this applies yet because you have not put AI in front of a real task, the sequencing question comes before the pricing one, and we laid out a slower starting path in our 30-day guide to using AI in a small business.
Frequently Asked Questions
How much does a ChatGPT Business Premium seat cost?
A Premium seat costs $125 per user per month, or $100 per user per month if you bill annually. Standard seats remain at $25 per user per month, or $20 billed annually. Premium includes five times the usage of a Standard seat and is not subject to the five-hour usage limit, with usage resetting weekly on both tiers.
Do I have to upgrade my whole team to Premium?
No. Workspace owners and administrators can mix Standard and Premium seats in the same workspace and reassign them as workloads change. For most small teams the sensible approach is to upgrade only the one or two people who repeatedly hit the usage ceiling, and leave everyone else on Standard seats.
What is the five-hour limit on ChatGPT Business?
It is a usage cap on Standard seats that stops a user once they have consumed a set amount of capacity in a rolling window, which interrupts long agent-driven tasks partway through. OpenAI removed this cap on July 12, reinstated it on July 31, and from August 10 made removing it a feature of the paid Premium tier.
Is ChatGPT Business Premium worth it for a small business?
It depends entirely on whether anyone on your team is being cut off mid-task, and you can answer that by observation rather than estimation. If nobody is hitting the limit, Premium buys you nothing you are currently short of. If one person is consistently blocked on long automated jobs, the upgrade costs $100 more per month for that person and removes the interruption.
One question we keep turning over: if the heaviest AI user in your business turns out to be someone nobody expected, what does that tell you about where your real workload has been hiding? Tell us in the comments.
ChatGPT Ads for Small Business: What You Can Buy
August 17, 2026 @ 2:41 pm
[…] this. If you sell to consumers making everyday decisions, the overlap is much better. Our note on who actually needs a premium ChatGPT seat is a decent proxy for which of your buyers sit on paid […]