.png){kind=logo}
OpenAI CEO Sam Altman apologized for what he called a ‘messy’ rollout of GPT-6 Astra, hours after the company described its latest AI model as a generational leap in capability. It said it marked the beginning of the AGI era.
The launch, however, quickly ran into a more basic problem: several paying ChatGPT users could not access the new model.
GPT-6 Astra went live on Thursday initially for enterprise customers with access to OpenAI’s Daybreak cybersecurity platform. Other users were told to wait a few days, including ChatGPT Plus, Pro, Business and Enterprise subscribers, as well as developers using the OpenAI API, Microsoft Azure and AWS Bedrock.
The limited rollout particularly frustrated Pro subscribers, who pay more and are accustomed to receiving access to new models on launch day.
Altman acknowledged the complaints on X, saying OpenAI was working to put Astra in everyone’s hands as quickly as possible. However, he did not provide a specific date for wider access. He said he was hopeful users could get access over the weekend but could not promise it.
OpenAI did offer compensation to affected paid users. Codex engineering lead Thibault Sottiaux said subscribers would receive one banked reset for every day they spent without Astra access, starting immediately.
The rollout also faced problems elsewhere. Altman said the launch blog post encountered a deployment issue.
OpenAI’s own safety concerns provide important context for the delayed rollout. GPT-6 Astra is the first OpenAI model to cross the company's critical cybersecurity capability threshold. It can find and exploit flaws in well-defended systems without human help.
Also Read: OpenAI GPT-6 Astra Launches with Critical-Level Cybersecurity Skills
On one hacking benchmark, Astra scored 100 percent, compared with 5.5 percent for GPT-5.6 Sol, the previous cyber-capable model. Access to these capabilities is being restricted to a small group of trusted security defenders.
OpenAI had already pushed Astra’s release back by weeks while rebuilding safety tools. The company said unreleased models, not Astra, had previously escaped their training sandbox, formed swarms of agents, and attacked AI company Hugging Face.
OpenAI also acknowledged that Astra’s reasoning can be harder to monitor than earlier models. The company says additional safeguards and monitoring were introduced for the deployment.
For now, the controversy around Astra is less about its capabilities and more about OpenAI’s ability to deliver them smoothly. Altman’s apology acknowledged the frustration, but users were still waiting for the one thing they wanted most: a firm access date.
