OpenAI has temporarily stopped accepting new customers for its $200-per-month ChatGPT Pro plan because demand for GPT-6 Astra is putting heavy pressure on its infrastructure. The decision mainly affects people who wanted the most extensive regular ChatGPT subscription; existing Pro accounts, lower-priced plans, and the application programming interface (API), which lets other applications access models, remain available according to the reports. At the same time, OpenAI is launching new features for voice conversations, company data, and automated workflows.
What is behind the Astra shortage
GPT-6 Astra launched on September 3, 2026, and has been rolling out to Pro, Plus, Enterprise, and Business accounts. OpenAI presents the model as a significant advance in reasoning, coding, and computer use. That positioning likely added to demand, although claims that Astra begins an era of Artificial General Intelligence (AGI), meaning broadly capable AI at a human level, remain provider marketing rather than an independently established result.
On September 10, TechCrunch reported the pause in new Pro sign-ups. OpenAI product leader Tibo Sottiaux said that this plan puts the greatest strain on the company’s systems. OpenAI described the suspension as the smallest measure that could preserve broad access while maintaining service for existing users.
The lead report from t3n lists the affected subscription at $200 per month in the United States and €229 in Germany. The supplied sources give neither a Swiss price nor a definitive availability status for Switzerland. OpenAI has also not disclosed how many people were signing up each day or when it expects to reopen the Pro tier.
Go and Plus are expected to remain available, as is API access. The shortage therefore does not mean that ChatGPT as a whole is closed or that Astra is completely inaccessible. It shows that OpenAI can restrict particularly compute-intensive premium usage separately from less expensive accounts and technical access.
What Astra is supposed to deliver in practice
OpenAI supports its performance claims partly with accounts from companies already using Astra. According to an OpenAI case study about Perplexity, the answer-engine company lets the model write communications, modify software, and monitor production systems. One concrete use involves application testing: Astra builds a small test program, simulates responses from external services, and checks a workflow from beginning to end.
Perplexity says it needs to check the model’s work much less frequently than it did with earlier generations. That could matter to organizations because the AI does more than suggest text or code: it performs several stages of a task and tests the outcome. However, these claims come from OpenAI and a participating customer and have not been independently verified in the supplied material.
The report about Cognition’s coding assistant Devin makes a similar but shorter claim. Astra is supposed to improve Devin’s ability to test its own software work and demonstrate that the result functions. The providers hope engineers will need to review less code manually, but the summary supplies no measurements or documented error rates.
Astra’s mathematics results also need careful interpretation. The model initially ranked first on ErdosBench, which contains 226 open mathematics problems, with a score of 3.23. It solved 106 problems, including 43 completely, and disproved another 27, but a later update moved Fable 5.1 into first place, making the statement that Astra led the benchmark outdated.
Benchmark developer Przemek Chojecki characterized the improvement across different mathematical research skills as a solid but limited five to ten percent and said the test was far from saturated. According to The Decoder’s assessment, OpenAI had also chosen not to make mathematics a priority for Astra. The model therefore appears stronger than earlier systems, but the results do not establish that it can reliably automate open mathematical research.
What OpenAI is launching alongside Astra
While access to Pro is restricted, OpenAI is opening other capabilities. GPT-Live-1 is a voice model that can listen and speak at the same time in full-duplex mode. Instead of connecting separate speech recognition, language, and speech-output systems in sequence, one model handles the conversation continuously, with the goal of reducing delays and responding better to interruptions.
GPT-Live-1 is already used in ChatGPT and is now available to outside applications through an API. It costs $0.05 per minute. Yelp uses it for telephone reservations and, according to its chief technology officer, has seen better call handling—a practical example is a restaurant phone assistant that can listen, answer, and react when a caller interrupts without restarting the exchange.
In OpenAI’s own benchmarks, GPT-Live-1 scored 80.1 percent on full-duplex tests, compared with 45.4 percent for GPT-Realtime-2.1. Response time fell from 1.4 to 0.8 seconds, while tool-call accuracy rose from 60 to 87 percent. These are provider benchmarks, not a guarantee of equal performance with Swiss dialects, noisy telephone lines, or industry-specific conversations; the source says twelve new voices cover different accents, dialects, and languages but does not provide a complete list.
A second release is the Data agent in ChatGPT Work. This data assistant connects to approved company sources such as Amazon Redshift, Google BigQuery, Databricks, MongoDB, Snowflake, Google Drive, and SharePoint. You can ask about changes in everyday language, refine the analysis during a conversation, and turn the findings into interactive dashboards.
OpenAI gives the workplace example of a sales team investigating why revenue growth has slowed. Another example is a customer team identifying which issues threaten renewals among the company’s largest accounts and what should be fixed first. The assistant is designed to use the organization’s own terminology, metric definitions, and calculation rules, but it does not remove the need to check whether the data is complete and the conclusions are plausible.
Pros and Cons of OpenAI’s new offerings
Pros:
- Multi-stage work – According to the case studies, Astra can modify software, build tests, and examine complete workflows rather than only returning isolated answers.
- More natural voice interaction – GPT-Live-1 listens and speaks simultaneously, which can help phone assistants respond more quickly to interruptions.
- Direct access to data – The Data agent opens business analysis to employees who do not want to learn query languages or a separate analytics tool.
- Alternative access – Plus, Go, and the API remain available according to OpenAI even while new Pro subscriptions are suspended.
Cons:
- Unknown duration – OpenAI has not said when Pro subscriptions will reopen or disclosed the actual scale of new demand.
- Provider evidence – Many performance figures and practical reports come from OpenAI or participating companies and have not been independently confirmed.
- Ongoing cost – GPT-Live-1 costs $0.05 per minute, which can become a meaningful expense when conversations are numerous or lengthy.
- Data risks – The Data agent accesses approved company sources, so each organization must decide what may be connected and who may see the results.
What this means for you and Switzerland
If you are just getting started, the Astra launch does not mean you need the Pro plan immediately. A sensible first step is to use an account that remains available for a narrow task, such as summarizing a report or explaining an existing set of figures, and then compare the result with the original. Apply the same approach to voice features: test names, misunderstandings, and dialects in your actual setting instead of treating a general benchmark as proof for your use case.
As an advanced user, you can get more value by dividing work into stages that can be checked and by providing suitable sources and clear success criteria. In a company, the Data agent could first describe a change in sales, then narrow it down to the affected regions, and finally produce a dashboard that colleagues can share. Human approval remains sensible before sending communications or modifying production systems because the case studies do not demonstrate perfect reliability.
The sources provide no specific details about Swiss prices, availability, or support for the country’s national languages. OpenAI says its Habitat storage system spans almost 40 geographic regions and includes data-residency features, meaning controls over where data is stored, as well as encryption. That is not a commitment to a particular Swiss region, so companies, schools, and public agencies cannot infer a specific privacy or storage-location guarantee from it.
The scale of the infrastructure at least shows why capacity is a substantial concern rather than a problem affecting a small service. OpenAI says Habitat handles more than 70 million requests per second and over 500 petabytes of data for products used by more than one billion people each week. Those infrastructure figures published by OpenAI describe online storage and do not prove that this particular layer caused the Astra shortage; compute-heavy model execution and data storage are separate parts of the system.
The rush for Astra demonstrates that even an expensive AI subscription may not be available on demand. At the same time, GPT-Live-1, the Data agent, and the Astra case studies show OpenAI shifting from individual chat answers toward voice interaction, data access, and multi-stage tasks. The unresolved issues are independent evidence of quality, availability in Switzerland, and how long OpenAI will need to restrict access to its most extensive subscription.
Sources
- Nachfrage nach Astra zu hoch: OpenAI nimmt vorläufig keine Neukunden für ChatGPT Pro mehr an – t3n, 2026-09-11
- OpenAI puts Pro subscriptions on hold due to Astra demand – TechCrunch, 2026-09-10
- Perplexity trusts GPT-6 Astra with end-to-end systems – OpenAI, 2026-09-14
- Cognition helps Devin test its own work with GPT-6 Astra – OpenAI, 2026-09-11
- GPT-Live-1: OpenAIs Full-Duplex-Sprachmodell ist jetzt für Entwickler verfügbar – The Decoder, 2026-09-10
- GPT-6 Astra lässt Mathematiker durchpusten, weil OpenAI es so will – The Decoder, 2026-09-10
- Now everyone can put data to work – OpenAI, 2026-09-10
- Rapidly scaling online storage to serve over 1 billion ChatGPT users – OpenAI, 2026-09-11
Image: Christina Morillo via Pexels