On September 10, 2026, OpenAI temporarily suspended new sign-ups for ChatGPT Pro, its top-tier subscription plan for individuals and professionals priced at $200 per month. The move comes as computational demand for GPT-6 Astra, the company's newest flagship model announced on September 3, 2026, has surged beyond expectations, placing unprecedented strain on OpenAI's server infrastructure. Existing Pro subscribers can continue using the service without interruption, but new applications will be paused until infrastructure capacity catches up with demand.
The decision was announced on X by Thibault Sottiaux, OpenAI's Vice President of Product for ChatGPT and Codex. Sottiaux emphasized that ensuring existing users continue to have an excellent experience and uninterrupted access to Astra was the top priority, explaining that this led to the decision to halt new sign-ups specifically for the Pro plan, which places the heaviest load on the system.
How the Sudden Halt on $200 Pro Sign-Ups Unfolded
The situation traces back to the September 3, 2026 launch of GPT-6 Astra. Almost immediately after release, Astra saw rapid adoption among developers and researchers, and usage began putting significant strain on OpenAI's entire infrastructure.
Sottiaux had already flagged the mounting pressure a day before the suspension was announced. On September 9, he posted on X: "Demand for Astra has reached a truly unprecedented scale. We're doing everything we can to keep up, but this is on another level compared to any growth we've experienced before." He added that while maintaining a great experience for existing users would always be the top priority, continued strain could force the company to temporarily pause new sign-ups for the Pro plan.
Just one day after that warning, on September 10, those concerns materialized. Sottiaux formally announced: "To ensure existing users get a great experience and continued access to Astra, we're pausing new sign-ups for the $200-per-month Pro plan." He explained: "Since Pro is placing the greatest load on our systems, we've taken this minimal step to preserve access as broadly as possible."
Importantly, the suspension applies only to new sign-ups for the $200-per-month Pro plan. The $20-per-month Plus plan for general users, the lower-cost Go plan, the business-oriented ChatGPT Business and ChatGPT Enterprise plans, and the developer-facing API all remain open for new sign-ups and continue to be offered as usual. Existing Pro subscribers are not losing access or having their subscriptions canceled — only the door for new sign-ups has been closed. OpenAI says it is working to expand infrastructure capacity, but has not provided a specific timeline for when new Pro sign-ups will resume.
Why Load Concentrates on the $200 Pro Plan
Why did OpenAI specifically target the $200-per-month Pro plan for suspension, rather than the $20-per-month Plus plan, which has a far larger user base? The answer lies in the usage limits set for each tier and the asymmetric way users engage with them.
ChatGPT Pro, at ten times the monthly price of Plus, has become the primary bottleneck, as heavy users concentrated there drive extremely resource-intensive inference workloads.
The $20 Plus plan imposes strict caps on the number of messages users can send to high-performance models within a given time window. In contrast, the $200 Pro plan was designed as a professional-tier offering built around frequent, extended access to the most advanced models. Power users running complex, sustained reasoning tasks naturally gravitate toward Pro rather than Plus.
The difference in billing structure compared to the API is also significant. When accessing GPT-6 Astra (model ID: gpt-6-astra) via the API, standard pricing charges $10 per million input tokens and $50 per million output tokens on a pay-as-you-go basis. Because API revenue scales directly with the amount of compute consumed, infrastructure costs and revenue stay tightly linked for API users.
ChatGPT Pro, however, is a flat $200 per month. Converted using the API's output token pricing ($50 per million tokens), that flat fee becomes remarkably cheap for anyone consistently generating output equivalent to millions of tokens via the API. Yet it's not unusual for developers running agentic features or large-scale refactoring tasks daily to consume tens of millions of tokens' worth of compute in a single month. As users flocked to a plan that permitted heavy inference at a flat rate, per-subscription compute consumption vastly exceeded OpenAI's projections, straining its server capacity.
How GPT-6 Astra's Reasoning Architecture Strained Infrastructure
At the root of this infrastructure constraint lies a technical characteristic of GPT-6 Astra itself.
Unlike conventional conversational chat models, Astra is built around the ability to autonomously carry out complex, multi-step tasks. It features a massive 1.05-million-token context window and has been optimized for "Computer Use" — controlling screens and browsers — as well as specialized software engineering and scientific research workflows. According to benchmark results published by OpenAI, the model achieved 57.9% on the Terminal-Bench 4.0 software engineering benchmark and 72.6% on OSWorld 2.0, which measures practical task performance within OS environments.
What enables these capabilities is "test-time compute" — a form of inference-time scaling in which the model autonomously runs through extensive internal trial and error before producing an output. For a single instruction from a user, the model may internally generate an enormous number of reasoning steps, verifying tool execution results and revising code along the way. The compute consumed per exchange can be tens to hundreds of times greater than that of conventional, simple text generation.
OpenAI had already raised usage limits for Codex users as recently as August 2026. But once Astra launched on September 3, the complexity of tasks users ran escalated sharply. As advanced reasoning loops ran in parallel at scale, the GPU clusters in OpenAI's data centers hit the limits of their supply capacity. OpenAI President Greg Brockman described Astra at launch as "the dawn of the AGI era," but it is precisely that high degree of autonomy that has placed the greatest strain on the company's computing infrastructure.
A Different Kind of Bottleneck From 2023, and a Turning Point for Flat-Rate AI
This is not the first time OpenAI has suspended new sign-ups for a paid plan due to overwhelming demand.
In November 2023, immediately after unveiling GPT-4 Turbo at its first developer conference, DevDay, CEO Sam Altman temporarily paused new sign-ups for ChatGPT Plus ($20 per month). At the time, the surge reflected generative AI's boom spreading across society broadly — the core issue was sheer volume, with massive numbers of general users connecting simultaneously.
This year's suspension, however, is fundamentally different in nature. Sign-ups for Plus remain open; only the $200 Pro plan has been paused. This indicates that what pushed infrastructure to its limits wasn't simple traffic volume, but rather the density of compute consumed by a relatively small number of power users through autonomous agents and inference-time scaling — a qualitative shift toward far heavier workloads.
This episode highlights a broader business model challenge facing the generative AI industry. For chatbots that simply return text instantly, a flat-rate monthly subscription model could reasonably balance profitability against compute capacity. But as AI enters an agentic era — where models think autonomously for hours, build code, and operate virtual desktops — the range of compute a single user might demand widens dramatically.
Offering unlimited access to top-tier models under a flat-rate structure becomes increasingly difficult to sustain, both in terms of physical data center capacity and cost structure. How OpenAI expands its data center capacity going forward, and how it transitions toward more flexible quota designs that blend flat-rate and usage-based pricing, will determine whether it can operate its AI platform sustainably.