OpenAI fixes GPT-6 Astra quality issues and propagates a usage reset, affecting around 5,000 users

On September 12, 2026, OpenAI Codex lead Thibault "Tibo" Sottiaux posted on X that, after working with a subset of users, OpenAI had identified and fixed several issues affecting GPT-6 Astra and had propagated a usage reset. The reset went into effect at roughly 4:00 PM Beijing time on September 12. Tibo listed the fixes: skill triggers written for older models that were firing too often or were blocking the model from checking its own work; an opt-in context-management experiment that could cause early stoppage or replies that pulled from older messages, which has been disabled and is estimated to have affected roughly 4,000 to 5,000 users; and several misconfigured engines that were producing measurable quality drops on long-tail traffic routed through them and have been removed. He said overall experience should be noticeably better, with more consistent follow-through, better tracking of the user's latest messages, and improved self-checking during work in progress. He thanked users who shared published examples and worked directly with OpenAI to isolate the problems, and described the work as part of OpenAI's ongoing quality process. GPT-6 Astra is OpenAI's flagship reasoning model released on September 3 and has been rolling out to Plus, Pro, Business, Enterprise, and commercial accounts throughout the month. Earlier in the week, OpenAI paused new subscriptions to the $200-per-month Pro 20X plan amid unprecedented demand for Astra.