Reported Codex Flag Connects OpenAI’s Astra to GPT-6, but Release Is Still Unclear

OpenAI has said Astra is coming soon. A reported Codex Desktop feature-flag screenshot now offers the strongest public clue that the system may ship under the GPT-6 name—but it also highlights the gap between deployment preparation and a product users can actually access. Neither reported entry establishes availability.

The September 3 screenshot reportedly showed two previously undocumented Astra-related model identifiers in a Codex Desktop Statsig feature flag, including one that connects Astra to GPT-6. The second identifier added an “Aeon” suffix, but the material provides no description of its behavior, access terms, pricing, or role. Treating Aeon as a separate public model would go beyond the evidence.

Feature flags are used to stage configurations and manage limited access. An internal identifier can remain unused, change before launch, or refer to a controlled test rather than ordinary customer access. That makes the screenshot evidence that a Codex delivery path may be getting ready, not proof that Astra—or a GPT-6-branded version of it—has launched.

OpenAI’s public model catalog still identified GPT-5.6 Sol as its flagship model on September 3, while the company’s safety material called the coming system Astra. The reported configuration therefore supplies the GPT-6 connection; OpenAI has not publicly confirmed that branding in the material here.

The naming question matters less than the release conditions. OpenAI has classified Astra as the first model to reach the Critical cybersecurity capability threshold in its Preparedness Framework, a designation it says applies when a system can find and exploit previously unknown vulnerabilities with the necessary tools and access.

OpenAI said a Daybreak Blue configuration scored 100% on ExploitBench, and that Astra found two previously unknown vulnerabilities during testing and used them in an exploit chain. Those are company-reported evaluation results for a specific configuration, not a description of the default production system customers will receive.

The controls users may encounter

• OpenAI plans to give selected partners early access to Astra’s advanced cybersecurity capabilities through Daybreak Blue before broader defensive access expands.
• For ChatGPT and Codex, a monitor may ask users to review an action before it proceeds; API jobs would stop when the same monitoring intervenes.
• OpenAI warns that the monitor can slow, pause, or stop legitimate work, including tasks unrelated to cybersecurity and jobs that run for extended periods.

That last limitation is especially consequential for Codex, where the value of an agent often depends on completing longer jobs. The reported flag makes a staged release plausible, but the formal launch still needs to establish the public name, define any Aeon variant, identify eligible users, and show how monitoring changes real task completion.