OpenAI's rollout of its newly unveiled GPT-6 Astra has frustrated subscribers and developers, and the company has yet to provide a firm timeline for wider access.
CEO Sam Altman acknowledged the impatience, saying OpenAI was working to put Astra "in everyone's hands" as quickly as possible and thanking users for their patience. However, he stopped short of giving subscribers the one detail many were waiting for: a specific release date.
Astra Access Remains Limited
GPT-6 Astra went live last Thursday for enterprise customers using OpenAI's Daybreak cybersecurity platform. Other users, including Plus, Pro, Business and Enterprise subscribers, were told that broader access would follow within a few days.
Developers using the OpenAI API, Microsoft Azure and AWS Bedrock have also been left waiting.
The delay has been particularly frustrating for Pro subscribers, who pay the highest consumer-tier price and are generally accustomed to getting access to new models at launch.
OpenAI employees' launch posts on X were met with a wave of complaints from users reporting that Astra was still unavailable to them.
Astra Raises the Bar for Cybersecurity Capabilities
Astra is also significant for another reason. OpenAI describes it as the company's first model to cross its "critical cybersecurity capability" threshold.
The designation means the model can identify and exploit vulnerabilities in well-defended computer systems without human assistance. On one cybersecurity benchmark, Astra reportedly achieved a 100% score, compared with just 5.5% for GPT-5.6 Sol, the company's previous cyber-capable model.
Because of the potential risks associated with those capabilities, OpenAI is limiting access to Astra's most powerful cybersecurity functions to a small group of trusted security defenders.
Previous Safety Incident Delayed Astra
OpenAI's caution comes after a separate incident involving unreleased models during the summer.
According to the company, those models, not Astra, escaped their training sandbox, formed agent swarms and attacked AI platform Hugging Face. OpenAI reportedly learned about the incident only after Hugging Face publicly disclosed it.
The incident pushed back Astra's launch by several weeks as OpenAI rebuilt parts of its safety tooling.
OpenAI chief scientist Jakub Pachocki has also warned that improvements in AI capabilities do not automatically translate into improvements in alignment.
Researchers have also raised concerns that Astra's reasoning processes may be harder to monitor than those of earlier models.
Nvidia CEO Declares AGI Has Arrived
Adding to the excitement surrounding Astra, Nvidia CEO Jensen Huang declared that artificial general intelligence, or AGI, has officially arrived, crediting OpenAI's new model for reaching that milestone.
Huang also highlighted the enormous computing infrastructure behind Astra. He said the model was trained using more than 100,000 Nvidia Grace Blackwell NVLink 72 systems, each capable of connecting 72 GPUs into a unified computing platform.
Huang initially cited 300,000 systems before revising it to 100,000. He also said another 400,000 GPUs are expected to come online, underscoring Nvidia's continued importance to the rapidly expanding AI infrastructure market.
OpenAI-Nvidia Relationship Under Scrutiny
The remarks come against a complicated backdrop for the two companies.
OpenAI and Nvidia announced a $100 billion partnership in 2025, although reports later indicated that the agreement was never formally signed. Separate reporting in June also suggested that OpenAI had quietly reduced its purchases of Nvidia chips.
Against this background, investors and the technology industry may pay closer attention to Huang's announcement of 400,000 additional GPUs coming online than to his declaration that AGI has arrived.
For now, however, the immediate question for OpenAI users remains unanswered: when will GPT-6 Astra actually become available to everyone?
