The short answer: there is a real change, but define the metric
This is not just an unsourced rumor. In a September 29, 2026 X post, Tibo (@thsottiaux), whose profile identifies his work on Codex and ChatGPT at OpenAI, announced that Pro $200 would reopen to new subscribers with a revised usage calculation. He said its API-spend equivalent would be half that of the old plan.
"half the dollar in API spend compared to the old Pro $200 plan"
He also committed to keeping the five-hour limit removed and argued that improving models and lower API prices would let subscribers accomplish more work. That is the provider's explanation, not an independently verified result for every workload.
Our reading: a reduction in the stated API-equivalent budget is supported by that announcement. A universal 50% reduction in completed tasks, tokens, or every existing subscriber's immediately available allowance is not established by it. The post announces a change alongside reopening; it does not provide a complete account-by-account transition schedule.
Half the budget is not automatically half the work
Three quantities need to stay separate: the subscription's allowance, the cost of a particular workload, and the number of acceptable results it produces. A headline multiplier such as "20x" also needs a defined comparison plan and measurement window. It is not itself a fixed token balance.
Illustration, not actual plan amounts: imagine an API-equivalent budget of 100 units and a repeatable task costing two units. That supports 50 attempts. If the budget falls to 50 units but the same task now costs one unit, it still supports 50 attempts. If the task continues to cost two units, it supports only 25. If the cheaper model needs more retries, accepted results can fall even when attempt capacity stays unchanged.
This arithmetic explains how the provider's argument could work. It does not prove that it works for an Astra-heavy workflow, a long-context coding session, or a user who cannot substitute a cheaper model. A cheaper model's price is not evidence that every model, setting, and workload has become cheaper by the same proportion.
What the X reactions actually say
In a September 29 reply, cole murray (@colemurray) challenged the claim that users would do more with less available usage. That captures a reasonable concern: a promise about future productivity is difficult to compare with a visible reduction today. The reply itself is not a controlled before-and-after measurement.
Nicolas Zullo's September 30 post took a different approach, drafting how he would communicate the change with a more explicit explanation of plan positioning and a transition for current users. His text is proposed announcement wording, not an official notice. Its suggested transition terms and performance figures should not be republished as verified policy or benchmarks.
These are two examples from the discussion, not a representative survey. They show that the dispute is about both usable capacity and communication. Neither justifies claiming that all subscribers have the same experience or that everyone is moving to another provider.
Our view: efficiency gains do not remove the need for clear terms
Lower inference prices are welcome. A subscription need not permanently preserve an arbitrary dollar-equivalent figure if the service genuinely delivers more useful work. But the customer should be able to test that claim against something concrete.
Someone paying for reliable access to a particular model is not necessarily buying an interchangeable bundle of cheaper tasks. Extra product features may be valuable to another customer while doing little for that person's existing workflow. A benefit elsewhere in the bundle is not automatically a replacement for the capacity they use.
Our criticism is therefore straightforward: announce the reduction plainly, document the transition, and publish comparable examples. "More work over time" is an ambition. Until it is tied to reproducible workloads and acceptable outputs, it is not a substitute for a readable allowance policy.
What subscribers should be told
- Old and new allowance: the unit, measurement window, comparison baseline, and how consumption is calculated.
- Who is affected and when: new subscriptions, existing renewals, billing-channel differences, and any temporary protection or credits.
- Model-specific effects: examples that keep model, context, reasoning, speed, and task quality comparable instead of quietly switching to a cheaper model.
- Bundle additions: which features draw on the shared allowance, which do not, and whether exemptions are permanent or promotional.
Those are our recommendations, not promises that OpenAI has made. A subscriber should check their own notice and dashboard rather than infer a renewal date or grace period from someone else's social post.
Measure the impact without blaming every change on the quota
The current official OpenAI documentation lists Pro tiers at $100, $200, and $500 per month and says Pro currently has no five-hour limit. Its usage guidance explains that model, context, reasoning, tools, retrieval, and caching affect consumption; current limits and reset times are shown in the usage dashboard. A public estimate is not a guaranteed number of completed jobs.
The same page specifies that Fast consumes included allowance at 2.5 times Standard for the same model, while purchased credits use a different multiplier. That is a consumption multiplier, not a speed guarantee. Changing speed settings can change effective capacity, but it is not evidence that a particular subscriber's reported loss was caused by that setting.
- Save the applicable plan notice and record the date, billing tier, displayed reset time, and whether temporary credits or a reset are present.
- Repeat a small set of saved tasks with the same model, reasoning, speed mode, inputs, and acceptance criteria.
- Track allowance consumed, retries, review time, accepted outputs, and whether a limit stopped work. Avoid mixing unrelated concurrent jobs into the comparison.
- Compare several sessions within equivalent windows. Redact account identifiers and private project material before sharing evidence or contacting support.
This is a proposed evaluation method. We have not measured a private Pro account or demonstrated a universal task-capacity reduction. The dated subscription guide and API pricing guide provide separate background on the two billing systems.
The bottom line
Do not dismiss the controversy as imaginary: the announcement explicitly describes a lower API-spend equivalent. Equally, do not turn it into a claim that every customer's output must halve. The useful question is what capacity a subscriber keeps for their actual work, under clearly stated terms. The provider should make that answer easier to obtain.
Sources and scope
Sources checked on October 1, 2026. The X posts below were read directly on their public pages. X's search and full timelines required sign-in, so this is not an exhaustive review of the platform. Statements about policy, individual reactions, hypothetical arithmetic, and our editorial judgment are distinguished above.