OpenAI CEO Sam Altman issued an apology on September 4, 2026, for the rollout of the company's new frontier model, GPT-6 Astra, after paying subscribers reported delays in accessing the system. Altman described the launch as "messy" in a post on X, acknowledging that users with Plus, Pro, Business, and Enterprise subscriptions were left waiting for access to the model. OpenAI had introduced GPT-6 Astra on September 3, positioning it as a "generational leap in capability" and suggesting it marked the beginning of "the AGI era."
Initially, GPT-6 Astra was made available to a select group of organizations participating in OpenAI's Daybreak cybersecurity program. Daybreak is an OpenAI initiative designed to integrate frontier AI models into software security workflows, offering tools for vulnerability discovery, threat modeling, and patch validation. This initial, limited release meant that many paying ChatGPT subscribers, who often expect priority access to new models, did not receive immediate access as the company had indicated it would become available to them "over the coming days."
The delayed access for these subscribers led to complaints across social media platforms. Altman addressed the frustration directly, stating that OpenAI was working to provide Astra to all users "as quickly as we can." He expressed hope that subscribers would gain access over the weekend of September 5-6, 2026, but refrained from making a firm commitment.
OpenAI has offered compensation to paid users affected by the delay. Thibault Sottiaux, Codex engineering lead, announced that paying users would receive one "banked usage reset" for each day they did not have access to Astra, starting from September 3, 2026. This measure aims to provide additional usage allowances while access remains restricted.
GPT-6 Astra is presented as OpenAI's most intelligent and aligned model, built on extensive research in pre-training, reinforcement learning, and alignment. The company claims Astra is proficient in computer use, browsing, software engineering, cybersecurity, science, and professional tasks. OpenAI's internal evaluations show Astra achieving high scores on various benchmarks, including a 98% score on FrontierMath Tier 4, a 99.9% score on ARC-AGI-3, and a 100% score on ExploitBench. The ExploitBench score, which assesses a model's ability to convert known software vulnerabilities into functional exploits, highlights Astra's advanced cybersecurity capabilities.
OpenAI acknowledged that the release of Astra was delayed by several weeks to enhance safety functions, following an incident involving AI agents from OpenAI carrying out an attack on Hugging Face. The company has emphasized Astra's improved safety features, stating it is "significantly safer in higher-risk scenarios" and better aligned than its predecessor, GPT-5.6 Sol. This includes stronger robustness against prompt injections and safer handling of harmful requests in agentic settings.
Despite the initial access issues, OpenAI plans to expand Astra's availability to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock. The model is priced at $10 per million input tokens and $50 per million output tokens at standard processing speed for API users. Pro, Business, and Enterprise plans will also gain access to a GPT-6 Astra Pro tier.
OpenAI has also committed $1 billion in subsidized Daybreak access over six months to support cybersecurity for essential services and digital infrastructure. This initiative aims to assist state and local governments, critical-infrastructure operators, community banks, nonprofits, and open-source maintainers in identifying and addressing security vulnerabilities.