OpenAI plans to cut the API-spend equivalent of its $200 Pro plan. That's a bad deal

OpenAI plans to reopen sign-ups for its $200 Pro tier, which it paused on September 10. In a post on X, Thibault Sottiaux said the company is also changing how that tier's included usage is calculated. By his account, the revised plan works out to roughly half the old plan's dollar equivalent of API spending. That does not necessarily mean half as many tasks: OpenAI has not published old and new limits by model. The monthly price is staying at $200. Asking customers to accept that change on the promise they will get more work done strikes us as a bad deal.

Sottiaux's post does not announce a cut to the $100 Pro tier. For the $200 plan, it does not yet say when the new calculation takes effect or whether existing customers' renewals will change. A person planning their week around that allowance still cannot tell how many comparable jobs the new plan will cover.

Efficiency gains are not a published allowance

OpenAI argues that its models will become more efficient and its API prices will keep falling. GPT-6 Sol and Luna launched with API rates 50% below comparable promotional GPT-5.6 pricing. If a hypothetical task falls from $4 to $2 at API rates, 100 tasks have half the dollar value without turning into 50 tasks. That is why 'half the API-spend equivalent' does not automatically mean half the tasks across every model.

But the inverse matters too. If a model's price does not fall by the same proportion, a lower dollar-equivalent allowance may buy less of that model's work. A long agentic job can consume more than a short one even when both produce one result. Sottiaux is asking subscribers to bank on future gains without showing how this change affects the work they already do. Improvements may arrive. They are not a substitute for publishing the new limits before people pay for them.

He also promises not to bring back the five-hour cap for this plan. Being able to spend the weekly allowance when you need it is genuinely useful. Preserving that flexibility, however, does not increase the allowance. OpenAI's Codex plan guidance says usage varies with the model, task, context and tools. A single dollar comparison cannot tell a working subscriber what happens to their normal workload.

What happens to the 20x label?

OpenAI's current Pro help page still describes $100 Pro as 5x and $200 Pro as 20x the usage of Plus. Those labels compare tiers with Plus. Sottiaux's post compares old and new versions of one tier using API prices. The denominators are different. We cannot responsibly turn '20x' into '10x,' or '5x' into '2.5x,' without knowing what happens to Plus and the $100 plan. OpenAI should explain whether it will keep the 20x label, what that label measures, and how allowances change for new and existing $200 customers.

Asking for a model-by-model allowance table is not pedantic. Someone paying $200 a month needs to know whether a normal week of work will fit. OpenAI could show the same model, task and settings under the old and new limits. Instead, its early explanation pairs an API-price reference with a productivity claim the customer must test after the fact. The burden of finding out what changed has moved to the person paying the bill.

Subscriptions are not dead, but the flat-rate advantage is under pressure

Sottiaux says OpenAI wants the value of a subscription dollar and an API dollar to converge over time. It is fair to read that as a warning that the flat-rate discount could narrow. It is not proof that subscriptions are ending. OpenAI already lets eligible users buy flexible usage credits after exhausting included usage. Those ChatGPT credits are separate from API credits. Buying them does not mean moving your work to the API.

He also hints at features that will not draw down usage, without naming them. Until we know whether those extras help with the work people buy Pro to do, they cannot offset a quantified cut. His post does not confirm a rumored $500 tier or suggest that the unnamed extras will consume more allowance. He says the opposite about those extras. OpenAI's DevDay may provide answers, but a customer should not need to wait for a keynote to learn the practical limits of a $200 subscription.

Even a strong DevDay does not erase the question about the allowance. OpenAI should publish a comparison by model and task, say when the change takes effect and state whether existing renewals are affected.