Markets Expect OpenAI to Release Its Next AI Model Within Weeks
Prediction markets are increasingly pricing in an imminent launch of OpenAI’s next major artificial intelligence system. Despite the company recently saying that development had slowed because of cybersecurity concerns, traders still believe the model-known internally as Astra and widely expected to be GPT-6-could arrive before the end of September.
On Polymarket, contracts tied to Astra’s release date assign approximately a 59% probability to a launch by September 15. The estimated likelihood rises to about 72% by the end of the month. These figures indicate that market participants see the delay as a temporary adjustment rather than evidence that the project has been pushed far into the future.
OpenAI has warned about cybersecurity risks
The market’s optimism conflicts with OpenAI’s latest public statement. On August 7, the company said it had slowed Astra’s rollout while addressing risks connected with the model’s capabilities. The announcement followed internal preparedness evaluations that identified major progress in agentic coding and cybersecurity.
OpenAI’s language was notably cautious. According to the company’s safety assessment, researchers could not rule out the possibility that the system might develop “critical cyber capabilities.” In practical terms, that means the model could potentially automate or significantly assist with advanced digital attacks, vulnerability discovery, exploit development, or other sensitive cybersecurity tasks.
The classification is important because OpenAI’s preparedness framework treats certain capabilities as a trigger for additional safeguards. A system reaching a critical level in cybersecurity could theoretically identify weaknesses in software and networks faster than human experts, generate sophisticated malicious code, or support operations requiring a high degree of autonomy.
Why traders remain confident
The strong prediction-market odds may reflect expectations that OpenAI’s delay concerns deployment procedures rather than the model’s core development. A company can postpone public access while continuing internal testing, red-team exercises, infrastructure preparation, and safety work.
Another possibility is that OpenAI may introduce Astra in stages. Instead of offering the full system to everyone at once, the company could begin with a restricted preview for selected customers, researchers, or enterprise users. Limited access would allow OpenAI to monitor behavior, evaluate abuse patterns, and adjust safeguards before a broader release.
Markets may also be reacting to the pace of competition in the AI industry. OpenAI faces pressure from companies developing rival frontier models, while customers and investors expect frequent improvements in reasoning, coding, multimodal interaction, and autonomous task execution. A lengthy delay could give competitors more time to attract users and developers.
What GPT-6 could bring
Although OpenAI has not formally described Astra’s specifications, expectations surrounding a GPT-6-class system include stronger reasoning, more reliable long-context processing, improved software development abilities, and greater autonomy when completing multistep tasks.
The model could potentially manage projects rather than merely answer individual prompts. For example, an advanced agent might break down a business objective, research relevant information, write and test code, analyze results, and revise its approach with limited human intervention.
Improvements in cybersecurity are likely to be both beneficial and controversial. The same capabilities that help defenders locate vulnerabilities, analyze malware, and strengthen networks could also be misused by attackers. This dual-use problem is one of the main reasons OpenAI may be moving more cautiously.
A release does not necessarily mean a full rollout
Prediction markets often treat a release as a specific event, but the term can cover several different scenarios. OpenAI could announce the model without making it immediately available to the public. It could release a smaller or less capable version first, provide access through an API, or limit usage according to risk assessments.
The company might also separate capabilities by user category. Enterprise customers could receive access under stricter monitoring, while consumer availability might come later. High-risk tools, such as autonomous coding or advanced cyber assistance, could be restricted, rate-limited, or subjected to additional verification.
For this reason, even if Astra appears within weeks, the practical impact will depend on how OpenAI defines availability.
Safety testing may determine the final schedule
Before launching a frontier model, developers typically conduct evaluations involving hallucinations, harmful instructions, privacy, cyber abuse, autonomous behavior, and resistance to manipulation. These tests can reveal problems that are not apparent during ordinary product development.
A model that performs extremely well in coding may require new restrictions around executable code. A system that can browse, plan, and interact with external tools may need stronger controls against unauthorized actions. If Astra demonstrates unexpected abilities, OpenAI could delay specific features while releasing safer functions.
This makes the release timetable difficult to predict from the outside. The model may be technically finished while the surrounding safety systems remain incomplete.
Prediction markets are not official confirmation
The probabilities shown by prediction markets represent trader expectations, not an announcement from OpenAI. Such markets can be influenced by rumors, speculation, liquidity, and traders interpreting the same public information in similar ways. A high probability does not guarantee that the event will occur on schedule.
A sudden change in evaluation results, a security incident, regulatory pressure, or a decision to redesign access controls could quickly reduce the odds of a September release. Conversely, a limited preview or developer announcement could be enough for traders to classify the event as a launch.
What users and businesses should watch
The most meaningful signals will be changes to OpenAI’s product documentation, developer tools, access policies, and safety disclosures. Evidence of new infrastructure, model-selection options, or expanded enterprise testing could suggest that a launch is approaching.
Businesses should also avoid making critical plans based solely on the expected arrival of GPT-6. New models often require changes to software integrations, security reviews, data governance, and operating costs. Early performance may vary by task, and access conditions could change during the initial rollout.
For users, the likely result is a gradual introduction rather than an overnight transformation. OpenAI may first emphasize reasoning, coding, or productivity while keeping the most powerful autonomous and cybersecurity-related capabilities behind strict controls.
The central question is capability management
The debate is no longer only about whether OpenAI can build a more powerful model. It is also about whether the company can safely deploy one with advanced agentic and cyber-related abilities.
If Astra does arrive within the next few weeks, the announcement will probably reveal as much about OpenAI’s safety strategy as about the model itself. A controlled launch would indicate that the company is prioritizing monitoring and risk reduction. A broad release with few restrictions would signal a more aggressive approach to competition.
For now, traders are betting that OpenAI’s next frontier model is close. The company’s own warnings suggest that the final timeline will depend on whether its security evaluations can keep pace with the system’s growing capabilities.