이미지: METAL LAB 생성
Summary
- OpenAI launched a Premium Seat for ChatGPT Business priced at $125 per month ($100 with annual billing)
- The existing Standard Seat ($25/month) remains, while Premium offers 5x usage and no 5-hour cap
- OpenAI said the pricing change reflects growing token consumption driven by an increase in agentic tasks
- 프리미엄 좌석 가격
- 월 125달러(연간 결제 시 100달러)
- 스탠더드 좌석 가격
- 월 25달러(연간 결제 시 20달러), 변동 없음
- 프리미엄 사용량
- 스탠더드 대비 5배, 5시간 사용 제한 없음
- 사용량 초기화 주기
- 매주
- 운영 방식
- 한 워크스페이스 안에서 두 좌석 등급 혼용 가능
- 오픈AI 설명
- 팀들의 작업이 복잡해지며 에이전트형 AI가 더 많은 토큰을 소모하기 때문
ChatGPT Has Outgrown $25
OpenAI has added a new tier to its ChatGPT Business plan. It's called "Premium Seats," priced at $125 per month. Choosing annual billing brings that down to $100, but it's still five times the existing Standard Seat price ($25/month, or $20/month billed annually). Since the Standard rate remains unchanged, this move looks less like a price hike and more like the addition of a new option: "pay more, use more."
Five Times Instead of Unlimited
The difference between the two seat tiers is spelled out in numbers. Premium offers five times the usage of Standard, and removes the 5-hour usage cap that applies to Standard. In exchange, usage resets weekly. Admins can assign different seat tiers to different team members within a workspace, making it possible to move only the employees who repeatedly run heavy workloads over to Premium.
| Category | Standard Seat | Premium Seat |
|---|---|---|
| Monthly fee | $25 (annual: $20) | $125 (annual: $100) |
| Usage | Baseline 1x bar:20 | 5x bar:100 |
| 5-hour cap | Yes | No |
| Reset cycle | - | Weekly |
The Flat-Rate Plan's Expiration Date
OpenAI explained that this change reflects the growing complexity of the work teams are doing with ChatGPT. The Decoder summed it up this way: "Agent-based AI consumes far more tokens." Agentic AI refers to a mode of operation in which the system plans on its own and repeatedly calls tools to complete a task, without needing step-by-step human instruction. Unlike the traditional chatbot pattern of a single question followed by a single answer, this kind of work involves many internal computation cycles and consumes far more tokens. That creates a cost structure that a flat-rate plan struggles to sustain.
There's also the argument that the flat-rate plans major AI companies have offered so far were, from the start, loss-leading incentives. In other words, a price increase was inevitable sooner or later. Microsoft's move to replace some of the higher-priced OpenAI and Anthropic models in Copilot with its own lower-cost, in-house models can be seen as another attempt to solve the same problem from a different angle.
Model Prices Are Falling While Service Fees Rise
What's notable is that the two trends are moving in opposite directions. On August 9, Amazon Bedrock cut API prices for GPT-5.6 Luna by 80% and for GPT-5.6 Terra by 20%. GPT-5.6 Luna's price dropped to around $0.20 per million input tokens and $1.20 per million output tokens. The cost of calling the model directly is falling, even as the price of ChatGPT as a finished product service is rising. This can be read as a signal that the center of gravity in pricing is shifting from "the model itself" to "how long and how extensively that model is run for a given task." Even with the same model, a short answer to a simple question and an hours-long agentic task consume vastly different amounts of tokens.
So What Actually Changes
For business customers, there's now a new calculation to make: who gets which seat. Employees who only need simple Q&A can get by fine on Standard, but teams that write large volumes of code or automate long research and analysis tasks may need Premium. OpenAI's decision to allow mixed seat assignments appears designed with exactly this kind of use-case separation in mind. More broadly, this announcement is one scene in a larger shift: AI service pricing is moving away from a structure where everyone pays the same price for the same thing, toward one where cost depends on how much work is actually being asked of the system. As agentic AI becomes more deeply embedded in everyday work, this kind of usage-based pricing is likely to spread beyond ChatGPT to AI services more broadly.
