OpenAI has reopened new sign-ups for ChatGPT Pro 200, its $200-per-month plan. The price is the same as before, but most new subscribers will get a smaller usage allowance than the plan offered previously.
A transition measure lets certain existing subscribers and re-subscribers keep the old allowance until October 29, 2026. After that, they will move to the new allowance.
OpenAI's Thibault Sottiaux says that, converted to API pricing, the new allowance is roughly half that of the old Pro 200. He also argues that improvements in model efficiency and lower API prices mean users can get more work done than before.
The real impact isn't determined by the "allowance is halved" figure alone. How much this change matters depends heavily on which model you use and how much processing you need to reach the quality you want.
The $200 Pro plan reopens, with the old allowance available until October 29
OpenAI's latest Pro help page states that Pro 200 is once again accepting new subscriptions.
As of September 30, it is not a planned relaunch: the plan can already be purchased from the pricing page.
The monthly price remains $200, but except for those eligible to keep the old allowance, new Pro 200 subscriptions come with a smaller usage allowance than before.
Eligibility for the transition measure isn't determined simply by whether you previously subscribed to Pro.
Users whose Pro 200 plan was active at the point OpenAI set for eligibility, or during the seven days before it, can keep using the old allowance until October 29 as long as they maintain their Pro 200 subscription.
Even if a Pro 200 subscription ended within the seven days before that point, re-subscribing may still make a user eligible for the transition measure.
| Subscription status | Usage allowance applied | Monthly price |
|---|---|---|
| New subscribers not covered by the transition measure | New, smaller allowance | $200 |
| Eligible existing subscribers and re-subscribers | Old allowance until October 29, while Pro 200 is maintained | $200 |
| After the transition period ends | Move to the new allowance | $200 |
Source: OpenAI's Pro help page. Prices are the monthly amounts shown for the US and do not represent final billing amounts in Japan.
The ability to keep the old allowance is time-limited.
The official help page does not state the specific date and time used to determine eligibility. OpenAI says it will email eligible users with details, so you can't determine eligibility from your sign-up date alone.
Also, canceling doesn't end a subscription immediately. You can usually keep using Pro until the current billing period ends, so the day you schedule a cancellation and the day your Pro access actually ends should be considered separately.
What is halved is the allowance converted to API pricing
Ahead of the Pro 200 reopening, Sottiaux posted on X that under the new way of calculating usage, the allowance is about half that of the old Pro 200 when converted to API pricing.
The key point: the monthly price has not been cut in half, and API credits have not been cut in half.
Pro 200 still costs $200 per month. When Sottiaux says "half," he means the value of the usage included in the flat-rate plan when converted to what the same work would cost through the API.
The usage allowance included in ChatGPT subscription plans, ChatGPT credits you can buy separately, and pay-as-you-go OpenAI API billing are all separate systems.
According to OpenAI's explanation of credits, once you use up the allowance included in your plan for eligible features, you can keep going using credits you've purchased.
These credits are separate from API credits.
In other words, Sottiaux is converting subscription usage into API prices for comparison. He is not saying Pro 200 comes with a fixed API balance.
In addition, on Plus and Pro, eligible agent features may share the same usage allowance.
Codex and ChatGPT Work share an allowance, and on eligible accounts, using ChatGPT for Word, Excel, PowerPoint and similar tools draws from the same pool.
So even if the amount of work you do in Codex stays the same, using other supported features more will reduce your remaining available usage.
Different screens or apps do not necessarily have their own independent allowances.
No return of the 5-hour limit; the weekly allowance can be used in bursts
While the total amount available shrinks, the timing of use becomes more flexible than before.
Sottiaux says the new Pro 200 will not reintroduce the "5-hour limit." The approach is to let users spend their weekly allowance in concentrated bursts when they need it.
This is a separate change from the amount of usage itself.
For example, if a system could halt you temporarily for "using too much in the last five hours" even when plenty of your weekly allowance remained, long focused work sessions would be difficult.
Without a 5-hour limit, you can use more on the days you need it, within the overall weekly allowance.
However, the removal of short-term limits is not the same as having a larger total amount for the week.
Even if the former improves, the fact remains that the new Pro 200 allowance is smaller than before.
API unit prices have fallen sharply, but that doesn't guarantee the same workload
With GPT-6 Sol and GPT-6 Luna, announced by OpenAI on September 22, API unit prices were cut substantially from the previous generation.
Comparing OpenAI's official prices per million tokens gives the following.
| Model generation change | Input: old → new | Output: old → new | Price drop |
|---|---|---|---|
| GPT-5.6 Sol → GPT-6 Sol | $4 → $2 | $20 → $10 | 50% for both input and output |
| GPT-5.6 Luna → GPT-6 Luna | $0.20 → $0.10 | $1.20 → $0.50 | 50% for input, about 58.3% for output |
The GPT-5.6 figures are compared against the promotional prices at the time. Extra charges for caching, tool usage, long context, fast processing and the like are not included.
Luna's output price fell from $1.20 to $0.50, which works out to a cut of about 58.3%.
So while not every price was cut uniformly to 50%, the API unit prices of the main models have dropped significantly.
This is the reasoning behind Sottiaux's explanation that even if the allowance's API-equivalent value is halved, users can do more work than before.
In a simple example, even if an allowance that used to be worth $100 of API usage shrinks to $50 worth, if the same model's unit price is also halved, the number of tokens you can process stays the same.
Furthermore, if the new model can finish the same job with fewer tokens or fewer attempts than before, the amount of work you can process could even increase.
However, this calculation can't be applied directly to Pro usage limits.
OpenAI's Codex pricing materials also state that you cannot accurately estimate how many tasks a subscription can run from API token prices alone.
Actual consumption varies with factors such as:
- The model used
- The scale and complexity of the task
- The length of the conversation or work history
- The amount of reasoning
- Tool use
- Search
- Caching
Jobs that look similar do not necessarily consume the same amount.
Also, if a cheaper model fails to reach the quality you expected and you have to revise repeatedly, the total cost including that extra work goes up.
Conversely, if the new model can reach the required quality with less processing, the effective workload may grow by more than the price drop alone suggests.
Which of these happens can't be determined from the API price list alone.
For difficult work that requires a higher-end model such as GPT-6 Astra in particular, the benefit of the Sol and Luna price cuts can't simply be carried over.
So Sottiaux's statement that users can do more work than before should not be taken as a measured result guaranteeing more work than the old Pro 200 for every model and use case.
Pro now comes in three tiers: $100, $200 and $500
Alongside this change, ChatGPT Pro's pricing structure has become three-tiered.
OpenAI's latest help page describes Pro 100, Pro 200 and Pro 500 as follows.
| Plan | Monthly price | Usage allowance | Astra Ultrafast |
|---|---|---|---|
| Pro 100 | $100 | Smallest among Pro plans | Not supported |
| Pro 200 | $200 | Larger than Pro 100 | Not supported |
| Pro 500 | $500 | Largest of the three plans | Supported |
Pro 500 includes the largest usage allowance of the three Pro tiers, plus "Astra Ultrafast."
On the other hand, buying additional credits on Pro 100 or Pro 200 does not unlock Ultrafast at launch.
In other words, buying extra credits to increase how much you can use is different from moving up to a higher plan to gain access to a faster model execution mode.
Users who can keep the old Pro 200 allowance until October 29 also don't automatically get Pro 500 features. Their subscription remains Pro 200.
A flat-rate plan's value can no longer be measured by "how many API dollars" alone
Sottiaux says that in the long run, API prices will fall further, and for many users a system of buying only the additional usage they need will make more sense.
The direction is to narrow the gap between how much usage you get from $1 spent on the API and $1 spent on a flat-rate plan.
He also says he wants to avoid pricing designs that deliberately keep list API prices high to make subscriptions look like a bargain, while using discounts to lower effective prices.
This describes his thinking about future pricing design; it is not an announcement that ChatGPT's flat-rate plans will be discontinued.
In fact, Pro is currently being expanded into three tiers at $100, $200 and $500.
As a result, judging Pro's value by "how many dollars of API usage it converts to" alone is no longer enough.
You need to consider the available models, speed, features, how the allowance is used, and how much additional credits can make up for.
For example, at the same $200, if your use centers on GPT-6 Sol and the new model reduces the processing needed per job, the impact of the smaller allowance may be limited.
On the other hand, if your work centers on long sessions with Astra, the reduction in API-equivalent value may show up more readily as an actual drop in workload.
For eligible existing Pro 200 subscribers, the transition period through October 29 can also serve as a time to gauge how much their workload changes under the new terms.
To compare, it helps to run the same type of work with the same model and reasoning settings, and record not only how quickly the allowance drains but also how many revisions were needed to finish.
If you can maintain the same quality and workload with about the same amount of allowance as before, model efficiency gains have offset the smaller allowance.
Conversely, if the allowance is consumed heavily before the same job is done, the claim that you can do more work than before even with half the API-equivalent value doesn't hold for that use case.