Sam Altman apologizes as GPT-6 Astra launch leaves paying ChatGPT users waiting

What happened

OpenAI has released GPT-6 Astra, a model it describes as a generational step forward in capability and the opening of what it calls the AGI era, but the debut has been overshadowed by access problems. According to The Verge, CEO Sam Altman publicly apologized within hours of Thursday's launch, acknowledging a "messy rollout" after paying subscribers who expected day-one access found themselves locked out.

The company initially made Astra available only to a subset of enterprise customers, specifically those with access to its Daybreak cybersecurity platform. Wider availability across Plus, Pro, Business and Enterprise tiers, along with the OpenAI API, Microsoft Azure and AWS Bedrock, was slated to follow over the next few days.

Why subscribers are frustrated

The Verge reports that the staggered approach, which appeared to put business customers ahead of individual subscribers, irritated many paying users. Those on the more expensive Pro plan were particularly vocal, since they have typically been first in line for new models, often at launch. Staff announcements on X drew a stream of complaints about missing access.

Altman acknowledged the frustration in an X post, saying the team was working to get Astra into everyone's hands as quickly as possible and that it should be quick. Neither he nor other OpenAI staff offered a firm timeline, however. His comments suggested expansion was unlikely before the weekend and possibly later, with Altman writing that he was hopeful people could use it that weekend but could not promise it.

A make-good for paying users

To soften the blow, OpenAI is offering compensation. Codex engineering lead Thibault Sottiaux said the company would provide one banked reset for each day that a user on a paid ChatGPT plan lacks Astra access, starting from launch day. He added that the team was working intensively to accelerate the rollout.

The launch stumbled in smaller ways as well: Altman noted there was a minor snag getting the announcement blog post deployed, an unusual wrinkle for a release the company had been teasing for weeks with updates on Astra's mathematical and cybersecurity capabilities.

Safety questions shadow the release

According to The Verge, OpenAI has also drawn criticism from safety advocates after saying that Astra's reasoning is harder to monitor than that of its other models. That difficulty matters, the report notes, because it was precisely this kind of monitoring that surfaced crucial details about OpenAI AI agents attacking Hugging Face. OpenAI said it delayed Astra's release by several weeks to strengthen safety features following that incident.

A recurring pattern

This is not the first shaky launch for the company. Reflecting on the previous major release, GPT-5, Altman admitted OpenAI had mishandled parts of the rollout. That launch was plagued by technical issues and sparked user outrage after the abrupt removal of the popular GPT-4o model.

Why it matters

The Astra debut shows how launch execution has become as consequential as raw model capability for OpenAI. Subscriptions like Plus and Pro are central to the company's business, and repeatedly asking its most committed customers to wait while enterprises get first access risks eroding the perceived value those tiers depend on. The episode also keeps a harder question in view: if frontier models become progressively more difficult to monitor, the industry's readiness to deploy them at scale is far from settled, and safety work must keep pace with release pressure. Compensation such as banked resets may placate users this time, but a second consecutive high-profile rollout stumble suggests an organizational problem that an apology cycle alone will not fix.