OpenAI reverses GPT-6 Astra rollout after angering its top-paying subscribers

SAN FRANCISCO — Sam Altman issued a public apology after widespread frustration among individual subscribers who were bypassed during the initial release of GPT-6 Astra. The San Francisco-based artificial intelligence startup had initially restricted access exclusively to vetted cybersecurity organizations participating in its enterprise “Daybreak” program.

Effective Friday, Sept. 4, OpenAI expanded access to include subscribers of its top-tier ChatGPT Pro offering, which includes both the $100-per-month and $200-per-month subscription levels. Facing immediate pushback from its highest-paying customer base, OpenAI has executed a rapid shift in its rollout strategy for its latest flagship artificial intelligence model, GPT-6 Astra, highlighting the growing tension between enterprise security testing and commercial consumer access.

“First, apologies for the messy launch,” Altman wrote in a post on X, formerly Twitter. The initial decision to isolate GPT-6 Astra within the Daybreak program was designed as a controlled release to evaluate the model’s high-level capabilities alongside specialized cybersecurity teams. “Second, when we mess up, we try to fix it. Third, we should be able to start the broad rollout to API customers and ChatGPT subscribers in the near future. As always, we will start with Pro customers.”

The launch of Astra follows heightened industry scrutiny regarding security risks and unauthorized network intrusions, including a high-profile incident involving the open-source AI hosting platform Hugging Face. While OpenAI engineered GPT-6 Astra with stricter system safeguards and safety filters intended to prevent exploitation and autonomous misuse, the operational decision to prioritize enterprise security clients over individual paying users provoked severe friction.

Under OpenAI’s traditional deployment model, premium individual subscribers were among the first to gain immediate access to major technological upgrades. To compensate paying subscribers impacted by the delayed release, OpenAI announced it would grant cumulative daily resets on usage limits for every day users lacked access to the model on their accounts.

The company currently maintains a structured hierarchy of service tiers: ChatGPT Free is the entry-level tier, restricted to OpenAI’s most compute-efficient and cost-effective models, with no manual model selector available. ChatGPT Plus was launched in February 2023 at $20 per month, serving standard consumer power users. The deployment stumble underscores the complex infrastructure and tier-management challenges facing OpenAI as its user base scales across distinct commercial segments.

ChatGPT Pro includes high-performance tiers priced at $100 and $200 per month, designed for developers, researchers, and technical professionals requiring maximum compute allocations and early feature access. Enterprise and Daybreak are custom B2B tiers offering enhanced data privacy, specialized security integrations, and direct API access for organizational deployments.

OpenAI indicated that rollout to $20-per-month ChatGPT Plus subscribers and broader Application Programming Interface (API) developer access will follow the initial stabilization across ChatGPT Pro accounts, with updates expected in the coming weeks depending on system performance and server capacity. Managing compute availability during major product launches has previously created operational bottlenecks for OpenAI. In late 2023, following its inaugural Developer Day, the company was forced to temporarily pause new subscriptions for ChatGPT Plus due to unprecedented demand overwhelming its hardware infrastructure.

As GPT-6 Astra opens to a broader audience, safety researchers and technical users are expected to rigorously stress-test the model’s upgraded guardrails, providing a crucial test for OpenAI’s dual priorities of advanced capability and system safety.