OpenAI Rolls Out GPT-6 Astra to Paying Users After Acknowledging Delayed Launch

The company made GPT-6 Astra available to all paying users with qualifying plans on Friday, completing the initial phase of its distribution just 24 hours after the model's public debut. Sam Altman, OpenAI's chief executive, acknowledged the stumbling start in a post on X, writing, "First, sorry for the messy rollout."

By 6:30 p.m. Eastern on Friday, September 4, GPT-6 Astra had reached ChatGPT users with Plus and Business plans. Thibault Sottiaux, a senior technical staff member at OpenAI, confirmed the expanded distribution in his own X post: "OK nevermind, the team and Astra did a good job and our systems are more scalable than we anticipated. Astra is now rolled out to all Plus and Business users too. Hope you have a blast and let us know how it goes!."

Earlier that day, Altman had announced that GPT-6 Astra was accessible to Pro, Enterprise, and Business Premium users in Work/Codex, as well as through the API, with rollout to Plus and Business tiers to follow. Subscribers to the $8/month ChatGPT Go plan will not gain access to either GPT-6 Astra or GPT-5.6, according to the company's pricing documentation.

https://x.com/thsottiaux/status/2096002992046796932?ref_src=twsrc%5Etfw

The rollout delay and developer frustration

When OpenAI unveiled GPT-6 Astra on Thursday, many developers and API users found themselves unable to access the new model immediately. Altman's apology acknowledged that neither API customers nor ChatGPT subscribers had received broad access at that point, and he indicated that wider availability would begin "in the near future," prioritizing ChatGPT Pro users.

The situation created an awkward position for the developer community. OpenAI had already published comprehensive API documentation and pricing details for Astra, including its 1.05 million-token context window, support for up to 128,000 output tokens, and pricing of $10 per million input tokens and $50 per million output tokens. However, the actual API endpoint remained unavailable, leaving developers unable to validate the model's performance claims against their own requirements.

https://x.com/sama/status/2095678759651438887?ref_src=twsrc%5Etfw

Behind-the-scenes infrastructure challenges

Sottiaux provided additional context regarding the constraints OpenAI faced during the launch. He stated that "We are starting to release GPT-6 Astra, and we are doing it as carefully and quickly as possible," and noted that the process would span "a few days." He explained that the rollout involved bringing new systems and additional compute capacity online as access expanded.

In his explanation, Sottiaux emphasized that "many novel systems will operate at scale for the first time" during the rollout and that OpenAI was "bringing a lot of compute up." While this did not definitively establish that compute limitations caused the initial delay, it offered insight into the infrastructure work required to support widespread access. OpenAI's original announcement had indicated that Astra would reach ChatGPT Plus, Pro, Business, and Enterprise users, along with the OpenAI API, Microsoft Azure, and AWS Bedrock, "over the coming days."

Developers who obtained early API access have already discovered additional complexities. The Astra API introduces a new category of safety-related interruptions that resemble timeouts but function differently, a distinction that could significantly affect those building production systems on the model.

Compensation through banked resets

To partially offset the access delay, OpenAI introduced what it calls a "banked reset" for affected users. Sottiaux announced that paid ChatGPT subscribers would receive one banked reset for each day they remained without Astra access, beginning September 3. He added that the "team is moving mountains to give access as fast as we can."

Banked resets, which OpenAI has previously offered with ChatGPT Work and Codex, allow users to restore their usage after reaching a limit. They differ from additional API credits or permanent usage increases, instead functioning as resets that users can accumulate and deploy when needed. The distribution method for Astra—one reset per day of unavailability—represents a departure from OpenAI's prior practices. To date, OpenAI has not announced a comparable compensation scheme for API developers awaiting access.

This approach aligns with OpenAI's broader experimentation in pricing models. The company recently unveiled an outcome-based pricing structure that charges only when the model delivers correct results, suggesting that its overall pricing strategy continues to evolve.

What remains unclear

For developers, significant questions persist regarding API availability and potential usage constraints. The timing of broad API access and whether it will include stricter rate limits remain unanswered. Astra's persistent-agent capabilities add weight to these concerns for those building on the platform. OpenAI has not yet responded to inquiries about the rollout issues, the banked reset program, or the API timeline.