ChatGPT Business adds $125/seat tier as agentic AI devours token limits

OpenAI is done pretending flat-rate enterprise AI is sustainable. The company just introduced Premium Seats for ChatGPT Business at $125 per user per month — or $100 with annual billing — while keeping standard seats at $25 monthly. That’s a fivefold price gap, and it reflects a very real physics problem: agentic AI workflows burn through tokens at a rate that makes the old pricing look naive.

Premium users get five times the usage capacity and no five-hour cap. Usage resets weekly rather than daily, which changes the rhythm for teams running long-horizon agent tasks. Admins can mix and match seat types within the same workspace, so power users get the heavy compute while occasional users stay on the cheaper plan. OpenAI frames this as responding to teams tackling more complex work, but the subtext is clearer: the company can’t keep subsidizing inference at scale.

This was always coming. The major labs priced their enterprise tiers aggressively to win adoption, and anyone who’s watched cloud economics knew those margins wouldn’t hold once customers started running chains of reasoning calls. Each agent step spawns multiple model invocations. That adds up fast. Microsoft saw the same writing on the wall and went the opposite direction — swapping pricey OpenAI and Anthropic models for cheaper in-house alternatives in Copilot to keep its own pricing stable.

What’s interesting here isn’t the price itself — $125 per seat is still cheap compared to what a knowledge worker costs — but what it signals about OpenAI’s cost structure. If agentic usage is stressing the system enough to warrant a dedicated premium tier now, the real question is whether $125 actually covers their compute bill or just buys them time. Either way, the era of all-you-can-eat enterprise AI is ending, one token at a time.