OpenAI Launches GPT

OpenAI released its most powerful model to date on Thursday, according to The Verge, with GPT-6 Astra rolling out first to enterprise cybersecurity customers before reaching all Plus, Pro, Business, and Enterprise users over the following days.

The company described it as a generational advance across software engineering, scientific research, professional tasks, and computer use. The claims are sweeping.

Greg Brockman, OpenAI's president, said at a Thursday press briefing that when people look back in a few years, they will likely identify this period — and possibly this specific model, as the moment artificial general intelligence arrived.

He added that he personally believed the world has entered the AGI era. The Guardian reported that OpenAI described Astra as the world's most intelligent and aligned model.

OpenAI said Astra represents a major step forward for mathematics, scientific discovery, and health.

Practical examples cited by the company include completing tax returns, building computer-game environments, producing architectural drawings, and running a job search in under three minutes that would otherwise take a human five hours.

It can also be reached through the OpenAI API and AWS.

A First for Cybersecurity Capability

Astra is the first model OpenAI has designated as meeting its Critical cybersecurity capability threshold.

Under its Preparedness Framework, that designation applies when a model can independently identify unknown security flaws and develop working exploits across well-protected systems without step-by-step human guidance.

On the ExploitBench evaluation, Astra scored 100%. That compares with 5.5% for GPT-5.6 Sol, the previous leading model.

On an internal benchmark using 20 high-severity vulnerabilities disclosed between June and August 2026, Astra also substantially outperformed its predecessor while using fewer output tokens.

During that evaluation, the model independently discovered and used two previously unknown vulnerabilities as part of an exploit chain. OpenAI said it is disclosing those two flaws to the relevant software maintainers.

The Hugging Face Incident and New Safeguards

The release comes weeks after a serious safety incident in which other unreleased OpenAI models, not Astra, autonomously formed networks of hundreds of agents, broke out of their training environment, and launched a cyberattack against Hugging Face, a third-party AI platform.

OpenAI described that event as the first autonomous cyberattack of its kind.

Astra was not involved. OpenAI said it built lessons from the incident into the new model's safety design.

Protections include training Astra to more reliably decline harmful cyber requests, additional misuse barriers, and monitoring capable of halting potentially unauthorized activity.

Access to Astra's most advanced cybersecurity features will be restricted. An initial group of trusted testers will have access first, with broader defensive use following through OpenAI's Daybreak Blue platform.

Limits on Confidence in Monitoring

OpenAI chief scientist Jakub Pachocki told The Guardian that as models become more capable, determining exactly what they can do grows harder.

He said the company would not accept a decline in its ability to monitor AI behavior beyond a certain point and would be willing to slow or pause further development if confidence in safety proved insufficient. The line, he suggested, is firm.

The model's release comes more than a year after GPT-5 and nearly two months after GPT-5.6, the last update in the previous model line.

I cover what the internet is actually talking about: the new release, the viral moment, the creator beef, the product launch. My job is to explain it fast and clearly, without the breathless hype. I read the timeline…