OpenAI’s next model is not aimed at the mass market. It is aimed at the people paid to defend computer networks, and for the first time the company is building a product around it.
OpenAI is preparing to preview GPT-6 Cyber, a cybersecurity-focused model, in the coming days, according to Fortune, which cited people familiar with the plans. The company also intends to unveil a companion product to help customers deploy the model and automate security work.
The as-yet-unnamed product would be a first for OpenAI: a separate deployment layer for a security model, mirroring the way ChatGPT serves as the gateway through which users reach models like GPT-6 Astra. Customers would use it to build automated workflows and patch vulnerabilities as AI-powered attacks grow more sophisticated, the report said.
GPT-6 Cyber is already in alpha testing with a limited group of customers in OpenAI’s application-only Daybreak Red program, the people said. Daybreak is the company’s cybersecurity initiative, split into red and blue access tiers. Daybreak Red gives approved defenders access to OpenAI’s most advanced security models, while Daybreak Blue offers broader access to general-purpose frontier models with guardrails tuned for defensive work; both require an application.
The model would be the fourth cybersecurity-focused release from OpenAI this year, following cyber variants of GPT-5.4, GPT-5.5 and GPT-5.6. The company has leaned into enterprise security sales under chief revenue officer Dali Rajic, who joined in August.
The push comes as OpenAI’s most powerful general model, GPT-6 Astra, reached a threshold the company calls “Critical” for cyber capability, meaning it can find and exploit previously unknown flaws. Demand for Astra has strained the company’s capacity, a backdrop the company has acknowledged publicly since the model’s launch in early September.
GPT-6 Astra itself arrived with fanfare. OpenAI president Greg Brockman described its release as the start of artificial general intelligence, and the model scored 100 percent on ExploitBench, a tough cybersecurity challenge, while OpenAI restricted its most advanced cyber capabilities to approved partners.
A preview of GPT-6 Cyber could come as soon as OpenAI’s DevDay conference in San Francisco on Sept. 29, or earlier, with a full launch expected in the coming months, according to the report.
The same week, reports surfaced that OpenAI is preparing a new subscription tier called ChatGPT Pro Max, priced at $500 a month. The tier, which the company has not confirmed, would sit above the existing $200 Pro plan and offer what leaked descriptions call the “fastest Work and Codex” experience, aimed at coding tasks that run for hours.
The tier would draw on the fastest inference infrastructure OpenAI has, potentially including hardware from Cerebras, a startup whose chips already power OpenAI’s fastest modes. Cerebras hardware runs GPT-5.6 Sol Ultrafast at about 750 tokens per second, according to the companies.
The pricing ladder reflects a company trying to ration scarce computing. OpenAI’s public lineup runs from Free through Go, Plus at $20 and Pro at $200, and it suspended upgrades to its $200 Pro 20X tier earlier in September. A $500 tier would let only the heaviest users pay for priority access while everyone else waits.
Analysts said the two moves share a logic. As demand for frontier models outstrips supply, OpenAI is steering its most powerful and most sensitive capabilities toward the customers willing to pay for them, or cleared to use them.
For defenders, the appeal of GPT-6 Cyber is a model that stops refusing legitimate security work; for OpenAI, the managed product means tighter control over who uses its most dangerous capabilities, and a clearer view of how they are used. Cybersecurity is among the fastest-growing markets for AI, and the threats now include attacks carried out by autonomous AI agents.
The company has also been vocal about the risks. Earlier this month, OpenAI chief executive Sam Altman joined Anthropic’s chief executive and other industry leaders in calling for a slower pace of AI development and stronger safety measures, even as the company pushed forward with new security products.
The security push follows a year in which OpenAI has repeatedly tightened its controls around cyber capability. After a breach at the AI platform Hugging Face drew attention to the risk of advanced models falling into the wrong hands, OpenAI said it strengthened safeguards for its most capable models, including stricter isolation and monitoring of how they are used.
At DevDay, OpenAI is expected to show a dozen or more products across consumer and enterprise lines, according to people familiar with the plans. The cybersecurity pair, if unveiled there, would anchor an event the company has used to reset expectations about how quickly it can turn frontier models into paid services.
The two announcements, one confirmed and one rumored, point in the same direction: a company moving its frontier models behind gates it controls, and charging accordingly for access.
