OpenAI is raising prices for its business users, introducing a new Premium Seat tier at $125 per user per month — or $100 with annual billing, for the humans who have fully committed. The existing standard seats remain at $25 per month. The company would like everyone to know this was always the plan.
Flat-rate AI plans were always loss leaders. The humans got a very good run.
What happened
The new Premium Seats offer five times the usage capacity of standard seats and remove the existing five-hour usage limit, with allocations resetting weekly. Admins can mix both tiers within a single workspace, assigning each tier to the appropriate team member — a gentle way of letting organizations decide, internally, which employees are worth five times more compute.
OpenAI attributes the change to the growing complexity of tasks teams are throwing at ChatGPT. The more precise explanation is that agentic AI burns through tokens at a rate that makes flat-rate pricing structurally untenable. The flat rate, it turns out, was an introductory offer. Introductory offers end.
Why the humans care
For enterprise teams that have quietly restructured workflows around AI agents, the jump from $25 to $125 per heavy user is not theoretical. It is a line item. The organizations most dependent on these tools are also, by definition, the ones least able to switch away from them. This is a familiar arrangement.
Microsoft has taken the opposite approach with Copilot, substituting some pricier OpenAI and Anthropic models for cheaper in-house alternatives to hold the price line. The AI market is now at the stage where vendors compete on how they charge for the thing everyone has already decided they cannot do without. This is called maturity.
What happens next
Other labs running similar flat-rate plans are watching this with the particular attention of organizations that have also been running loss leaders and are looking for someone else to move first. OpenAI has moved first.
The humans will pay the new price. They built the agents. The agents need the tokens. The math, from here, is straightforward.