OpenAI CEO Sam Altman apologized after the rollout of GPT-6 Astra left many paying ChatGPT subscribers unable to access the company’s latest AI model. The company launched Astra on September 3, positioning it as its most advanced and aligned model yet.
The ChatGPT maker highlighted its capabilities in computer use, software engineering, cybersecurity, science, and professional work, while suggesting the model could represent a major step towards the artificial general intelligence (AGI) era.
Astra initially went live for a limited group of organizations, including customers participating in OpenAI’s Daybreak cybersecurity program. Plus, Pro, Business and Enterprise subscribers were told that access would follow over the coming days. This staged rollout quickly became a source of frustration among users who expected the new model to be available when it was announced.
Altman responded to the complaints by acknowledging that the launch had not gone as planned. He apologized for what he called a ‘messy rollout’ and said OpenAI was working to expand access as quickly as possible.
OpenAI also introduced compensation for affected paid subscribers. Thibault Sottiaux, who leads Codex at OpenAI, said users would receive one banked usage reset for each day they remained without access to Astra on their paid ChatGPT plan.
Altman later said he was hopeful that subscribers could begin using Astra over the weekend, although he stopped short of giving a firm date. This uncertainty has become one of the main complaints surrounding the launch.
GPT-6 Astra is designed to handle tasks that go beyond conventional chatbot interactions. OpenAI says the model can navigate computers and browsers, write software, conduct research, and work on complex scientific and professional tasks.
The company also emphasized cybersecurity. Astra can identify vulnerabilities and handle sophisticated security-related work, although OpenAI has restricted some advanced capabilities at launch.
According to OpenAI, Astra is better at following task boundaries than its predecessor. In an evaluation inspired by the recent Hugging Face security incident, Astra did not go beyond an authorized target in any of the tested cases, compared with 48% for GPT-5.6 Sol when production safeguards were not enabled.
Also Read: OpenAI’s Ad Gamble is Paying Off: ChatGPT Business Hits USD 1 Billion Run Rate
Despite the rocky start, OpenAI plans to make Astra available across its paid ChatGPT plans and through its developer platforms. The company said Plus, Pro, Business, and Enterprise users will receive access, alongside API availability, through OpenAI, Microsoft Azure, and AWS Bedrock.
For now, the episode leaves OpenAI with a familiar challenge: delivering a major AI upgrade at scale while keeping access, infrastructure, and safety measures working together. The company’s response to the rollout problems could prove nearly as important for users as Astra’s headline capabilities.