OpenAI's $200-a-month Pro plan is back on sale, but it does not buy what it used to. Starting September 30, new subscribers pay the same price but get roughly half as much usage as the plan offered a month ago, measured against what that work would cost if bought directly through OpenAI's pay-per-request pricing.
The timing explains why. OpenAI had paused new Pro sign-ups on September 10 after demand for its newest and most capable model, GPT-6 Astra, overwhelmed its computers. Rather than simply raising the price to manage demand, OpenAI recalculated how much usage $200 buys.
OpenAI's argument is that subscribers will not actually notice a loss. This week the company also launched two cheaper models, GPT-6 Sol and GPT-6 Luna, at half the price of the models they replaced. If you do your work on these newer, cheaper models, the math nets out close to even. If your work depends on Astra, the flagship model that did not get a price cut, you are simply getting half as much for your money.
That split matters. Astra is the model OpenAI has marketed as its most capable, the one businesses reach for on their hardest tasks: complex coding, detailed research, multi-step projects. Those are exactly the users a $200-a-month plan was built to attract, and they are the ones absorbing the full cut.
One small win for subscribers: OpenAI confirmed it will not reinstate a five-hour usage clock that used to force people to space out their work. Regular ChatGPT Plus subscribers got that clock reimposed in August because casual users kept accidentally burning through a full week's allowance in one sitting. Pro subscribers keep the freedom to use their weekly allowance whenever they want.
The bigger story here is not one plan's pricing. It is where AI subscriptions are headed generally. For years, companies like OpenAI sold flat monthly access to expensive computing power at a loss, using investor money to make AI look cheap and get people hooked. That subsidy is thinning out. Microsoft already runs parts of its Copilot product on a pay-as-you-go credit system rather than a flat fee. Anthropic has started billing its most capable Claude models to subscribers through usage credits tied to its own developer pricing, the same mechanic OpenAI just applied to Pro.
For any business that has budgeted a flat monthly cost for AI tools, this is worth watching closely. The flat-fee era of AI subscriptions was never going to last, because the computing costs behind these tools do not shrink to zero just because a company wants to sell a simple monthly price. As models keep getting more capable and more expensive to run at the high end, expect subscriptions everywhere to look less like a fixed software license and more like a utility bill that moves with how much you actually use.
The practical takeaway is simple. Treat AI spending as a variable cost to monitor, not a fixed line item to set and forget. Track which employees or teams use the most expensive models for the heaviest tasks, because those are the users most exposed to price changes like this one. Businesses that build in usage tracking now will handle the next round of price changes far more smoothly than those who assume this month's bill will look like next month's.