Astra is so great, even OpenAI can't keep up.
Early Friday morning, OpenAI announced a pause on new subscriptions and upgrades to the ChatGPT Pro $200 (Pro 20X) plan; existing users are unaffected.
As Tibo (Thibault Sottiaux, Head of Core Products and Platforms at OpenAI and former lead of Codex) stated, the Astra model's exceptional performance has generated unprecedented demand, and there is currently insufficient computing power to ensure a smooth user experience.
The new supply and demand dynamics are unpredictable; OpenAI has not provided a specific timeline for resuming sales, only stating that it is working to increase capacity as quickly as possible.
This matter has been confirmed by OpenAI's Help Center. It must be said that recent temporary announcements from large model companies have been coming in rapid succession, making it hard to keep up.
Lobster founder Peter Steinberger also said it's best to subscribe soon.
Astra indeed consumes credits faster than GPT-5.6 Sol; OpenAI has previously warned users that Astra depletes Work/Codex quotas more rapidly, with a pay-per-use rate approximately 2.5 times that of Sol. Additionally, the most powerful Pro version of GPT-6 Astra is exclusively available to Pro, Business, and Enterprise users, with Pro 5X offering 50 credits per week and Pro 20X offering 200 credits per week.
On September 3, OpenAI released GPT-6 Astra, positioning it as the "most powerful AI," and the official use of the term "AGI era" sparked intense discussion.
The users who have switched to the new model best illustrate its capabilities; many in the community had already claimed that OpenAI instantly poached Anthropic’s users. OpenAI also seems to welcome this narrative—on September 9, Tibo posted a warning stating, “Demand for Astra has never been higher; we’re pulling every lever we can… If this continues, we may have to pause new Pro subscriptions.”
It was unexpected that the most popular AI company would implement rate limiting on its core product, sparking dissatisfaction among developers.
We know that ChatGPT has subscription tiers such as Go, Plus, and Pro 5X—why did the most expensive one sell out first?
It appears that, within OpenAI’s current pricing structure, the Pro 20X plan—though the most expensive and least subscribed—offers the lowest cost per unit of compute and the highest utilization rate. For OpenAI, it may paradoxically be the most stressful tier. Upgrading from Pro 5X to Pro 20X costs users twice as much but provides four times the quota, and Pro 20X also includes unlimited desktop voice and other benefits. Thus, the most expensive plan indeed offers the greatest discount.
SemiAnalysis calculated that, if you exhaust the weekly limits for each tier of OpenAI and Anthropic subscriptions and then convert to standard API pricing, OpenAI begins to lose money at usage rates above 11.4% for Plus and Pro 5x, and reaches negative gross margins at just 5.7% usage for Pro 20x (Anthropic is around 10%). This does not even account for the reset cards occasionally issued by Tibo.
Moreover, the people who actually buy Pro 20X are the ones who will truly use it up—typically developers running long-chain Agent tasks, keeping Codex running all night, and dispatching hundreds of tasks per day.
No wonder people complain—subscribing to the Coding Plan feels like winding yourself up, just thinking about working overtime.
On the other hand, the sell-out of Pro 20X indicates that companies like OpenAI originally assumed that most users would not fully utilize their Coding Plan; however, after Astra's launch, the actual usage patterns of heavy users have surpassed this assumption.
As top-tier AI applications become increasingly useful, their prices are likely to continue rising.
Reference content:
https://help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers
This article is from the WeChat public account "Machine Heart" (ID: almosthuman2014), authored by Machine Heart focusing on large models, and edited by the Machine Heart editorial team.