The end of "unlimited."
ChatGPT adds a $125 premium seat.
OpenAI added a "Premium Seat" tier to ChatGPT Business — $125/month, five times the usage of the standard seat. As agentic AI burns through tokens fast, the flat-rate "unlimited" seat is starting to get tiered.
Agentic AI is starting to
outpace the flat-rate plan
The premise of "flat monthly fee, unlimited use" is buckling under the real cost of running AI agents.
OpenAI added a Premium Seat tier to ChatGPT Business at $125/month ($100/month billed annually). Compared to the Standard seat ($25/month, $20 annually), Premium gets 5x the usage and removes the five-hour usage cap that applies to Standard seats. Per OpenAI's official announcement, Standard and Premium seats can be freely mixed, person by person, within the same workspace.
The backdrop, per THE DECODER's reporting, is a real operational friction: agentic AI burns through tokens fast, and heavy users on the Standard plan were hitting the five-hour cap often. As usage shifts from one-off Q&A to multi-step, semi-autonomous agent tasks, the variance in how much any one user consumes has widened — making a flat "everyone gets unlimited" price hard to sustain.
What matters in the design is that OpenAI isn't asking for an all-or-nothing upgrade. Workspace admins can assign seats per person, so in practice a team can move just the small number of users who actually hit the cap onto Premium while leaving most people on the $20-25 Standard seat. As a sign-up incentive, early adopters also get $100 in workspace credit per Premium seat, up to $500 for five seats.
The two seats, compared
| Standard Seat | Premium Seat |
|---|---|
| $25/mo ($20 annual) | $125/mo ($100 annual) |
| Capped every 5 hours | No usage cap |
| Baseline usage | 5x baseline usage |
| Mixable in one workspace | Mixable in one workspace |
Who it affects, and how
Heavy users & developers
If you kept hitting the five-hour cap on long agent runs, this removes real friction. Whether the extra $100-125/month is worth it depends entirely on your actual usage.
Light users
Basically unaffected. Standard pricing and features stay the same, and only the people who need it get switched to Premium.
IT & procurement
Built for mixed deployment, not a blanket switch. Picking who goes Premium becomes its own ongoing task — a 50-person org that upgraded everyone would add over $60,000/year, so decisions need to be usage-log driven.
Unlimited
just stopped being uniform.
What "5x" actually measures is unclear
The fact that OpenAI felt it needed a sign-up incentive to smooth the transition is itself telling — this is a cost-structure shift real enough to need cushioning. There's also a transparency question: what exactly does "5x usage" measure — tokens, messages, tasks? Public materials don't say. Agent-task costs vary a lot, so how much real headroom you actually get depends heavily on how you use it.
What to watch next is whether other major AI vendors follow with their own heavy-user tier. Since agentic AI's compute cost is an industry-wide problem, this kind of price tiering could become the norm rather than the exception. For any org considering the switch, the practical move is to check a few weeks of usage logs to see who's actually hitting the cap before assigning seats — that's how to balance the cost against the benefit.